{ "site": "https://huecki.com/", "generatedAt": "2026-08-23T12:54:41.119Z", "indexes": { "llms": "https://huecki.com/llms.txt", "llmsFull": "https://huecki.com/llms-full.txt", "rss": "https://huecki.com/rss.xml", "sitemap": "https://huecki.com/sitemap-index.xml" }, "markdownConvention": { "blog": "/en/blog/{slug}.md and /blog/{slug}.md", "tools": "/en/tools/{slug}.md and /tools/{slug}.md" }, "topics": [ { "id": "agent-harnesses", "title": "Agent Harnesses", "titleDe": "Agent Harnesses", "description": "Control layer around agents: phases, allowed actions, evidence, exit conditions, and review rules.", "descriptionDe": "Kontrollschicht rund um Agenten: Phasen, erlaubte Aktionen, Evidenz, Exit-Bedingungen und Review-Regeln.", "urls": { "en": "https://huecki.com/en/topics/agent-harnesses/", "de": "https://huecki.com/topics/agent-harnesses/" }, "entries": [ { "collection": "blog", "slug": "agent-eval-too-short-trajectory", "title": "Your Agent Eval Is Too Short", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-eval-too-short-trajectory/", "markdownUrl": "https://huecki.com/en/blog/agent-eval-too-short-trajectory.md", "summary": "A final pass/fail score hides the part of agent work that matters most: where the run started drifting, whether it noticed, and whether it recovered. The practical replacement is a trajectory eval with checkpoints, failure labels, and recovery metrics.", "tags": [ "AI Agents", "Evals", "Developer Workflow", "Agent Harness", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-harness-is-a-binary", "title": "Your Agent's Harness Is a Binary Now", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-harness-is-a-binary/", "markdownUrl": "https://huecki.com/en/blog/agent-harness-is-a-binary.md", "summary": "Two 2026 papers from the same research lineage quietly retire prompt engineering as a discipline. The agent's system prompt is now a binary you can version, diff, and evolve with a 200-line loop. The four metrics that actually matter are not the ones your dashboard shows.", "tags": [ "AI Agents", "Agent Harness", "Evals", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-harness-needs-a-behavior-map", "title": "Your Agent Harness Needs a Behavior Map", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-harness-needs-a-behavior-map/", "markdownUrl": "https://huecki.com/en/blog/agent-harness-needs-a-behavior-map.md", "summary": "Harness Handbook points at a practical bottleneck in agent engineering: the behavior you want to change is scattered across prompts, state managers, tool calls, policy code, and tests. Build a behavior map before editing the harness.", "tags": [ "AI Agents", "Agent Harness", "Coding Agents", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-is-not-the-product-fitness-function-is", "title": "Better AI Products Need Systems, Not One Agent", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-is-not-the-product-fitness-function-is/", "markdownUrl": "https://huecki.com/en/blog/agent-is-not-the-product-fitness-function-is.md", "summary": "Better AI products come from improvement systems around the agent. This guide shows how to build one with deterministic checks, narrow scoring rubrics, private holdouts, calibrated judges, and promotion gates.", "tags": [ "AI Agents", "Agent Evals", "Self-Improvement", "Coding Agents", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-state-machines-seatbelts", "title": "Give Your Agent Seatbelts, Not a Longer Prompt", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-state-machines-seatbelts/", "markdownUrl": "https://huecki.com/en/blog/agent-state-machines-seatbelts.md", "summary": "When an agent keeps jumping from planning to editing to testing at the wrong time, the fix is not usually another paragraph of system prompt. Put the workflow into explicit states, give each state a tiny tool policy, and make phase changes visible.", "tags": [ "AI Agents", "Coding Agents", "Agent Harness", "Developer Workflow", "AI Safety" ] }, { "collection": "blog", "slug": "agenten-brauchen-runtime-contracts", "title": "Agenten brauchen Runtime Contracts, nicht längere Prompts", "language": "de", "canonicalUrl": "https://huecki.com/blog/agenten-brauchen-runtime-contracts/", "markdownUrl": "https://huecki.com/blog/agenten-brauchen-runtime-contracts.md", "summary": "Bessere Prompts machen Agenten nicht automatisch zuverlässig. Entwickler brauchen Runtime Contracts: explizite Verträge dafür, welche Tools ein Agent nutzen darf, was er erinnern darf, wann er stoppen muss und wie seine Behauptungen geprüft werden.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Developer Workflow", "KI-Workflows", "Evals" ] }, { "collection": "blog", "slug": "agents-dont-need-longer-prompts-they-need-harnesses", "title": "Agents Don’t Need Longer Prompts. They Need Harnesses.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses/", "markdownUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses.md", "summary": "The arXiv survey Code as Agent Harness names the next shift in agent engineering: code is not only what agents generate. It is becoming the executable, inspectable, stateful runtime that makes agents reliable.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Coding Agents", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-coding-agent-harness", "title": "AGENTS.md reicht nicht: Dein Coding Agent braucht einen Harness", "language": "de", "canonicalUrl": "https://huecki.com/blog/agents-md-coding-agent-harness/", "markdownUrl": "https://huecki.com/blog/agents-md-coding-agent-harness.md", "summary": "Ein Coding Agent wird nicht durch einen magischen Prompt zuverlässig. Er braucht einen Harness: AGENTS.md, Skills, Tool-Permissions, Hooks und Evals, die merken, wenn sich sein Verhalten verschiebt.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-coding-agent-harness-en", "title": "AGENTS.md is not enough: your coding agent needs a harness", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en/", "markdownUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en.md", "summary": "A coding agent is not made reliable by one magic prompt. It needs a harness: AGENTS.md, skills, tool permissions, hooks, and evals that catch behavior drift.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-control-surface", "title": "AGENTS.md ist kein Kontext. Es ist eine Steuerfläche.", "language": "de", "canonicalUrl": "https://huecki.com/blog/agents-md-control-surface/", "markdownUrl": "https://huecki.com/blog/agents-md-control-surface.md", "summary": "Die überraschende Lektion aus AGENTS.md-Benchmarks ist nicht, dass Kontextdateien nutzlos sind. Sie verändern Agent-Verhalten, manchmal in Richtung teurerer und weniger nützlicher Arbeit. Behandle sie als Steuerfläche, nicht als Repo-Handbuch.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Context Engineering", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-control-surface-en", "title": "AGENTS.md Is Not Context. It Is a Control Surface.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-md-control-surface-en/", "markdownUrl": "https://huecki.com/en/blog/agents-md-control-surface-en.md", "summary": "The surprising lesson from AGENTS.md benchmarks is not that context files are useless. It is that they change agent behavior, sometimes into more expensive and less useful work. Treat them as a control surface, not a repo manual.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Context Engineering", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-need-exit-conditions", "title": "Agents Don’t Need ‘Keep Going’. They Need Exit Conditions.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-need-exit-conditions/", "markdownUrl": "https://huecki.com/en/blog/agents-need-exit-conditions.md", "summary": "The useful lesson behind Claude Code /goal is not that agents can run forever. It is that long-running agent work needs an explicit, observable exit condition: what proves done, what stays in scope, and when to stop blocked.", "tags": [ "AI Agents", "Agent Harness", "Developer Workflow", "AI Engineering", "Evals" ] }, { "collection": "blog", "slug": "ai-agents-need-evidence-before-clicking", "title": "AI Agents Need Evidence Before They Click", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-agents-need-evidence-before-clicking/", "markdownUrl": "https://huecki.com/en/blog/ai-agents-need-evidence-before-clicking.md", "summary": "When an agent clicks, sends, pays, deletes, or extracts data, the critical truth cannot live only in model prose. Put a small evidence gate before risky tool calls: predicate, evidence type, source, decision.", "tags": [ "AI Agents", "Multimodal AI", "Browser Agents", "AI Safety", "Developer Workflow" ] }, { "collection": "blog", "slug": "ai-slop-gate-after-tests-and-lint", "title": "Put an AI Slop Gate After Tests and Lint", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-slop-gate-after-tests-and-lint/", "markdownUrl": "https://huecki.com/en/blog/ai-slop-gate-after-tests-and-lint.md", "summary": "Tests tell you whether behavior still works. Linters tell you whether code is syntactically and stylistically acceptable. An AI-slop gate catches the residue coding agents leave behind: fake comments, swallowed errors, any-casts, duplicated helpers, TODO stubs, and dead code.", "tags": [ "AI Engineering", "Coding Agents", "Developer Workflow", "Code Quality", "Evals", "Agent Harness" ] }, { "collection": "blog", "slug": "coding-agent-instruktionen-testen", "title": "Teste deine Coding-Agent-Instruktionen wie Code", "language": "de", "canonicalUrl": "https://huecki.com/blog/coding-agent-instruktionen-testen/", "markdownUrl": "https://huecki.com/blog/coding-agent-instruktionen-testen.md", "summary": "Eine gute AGENTS.md ist kein Prompt-Dokument, das man einmal schreibt. Sie ist ein kleines Betriebshandbuch fuer den Agenten. Und Betriebshandbuecher werden besser, wenn man sie gegen konkrete Fehler testet.", "tags": [ "Coding Agents", "AI Engineering", "Developer Workflow", "AGENTS.md", "Evals" ] }, { "collection": "blog", "slug": "coding-agents-need-hardened-harness-evals", "title": "Coding Agents Need Hardened Harness Evals", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/coding-agents-need-hardened-harness-evals/", "markdownUrl": "https://huecki.com/en/blog/coding-agents-need-hardened-harness-evals.md", "summary": "Permissive coding-agent benchmarks hide a boring production truth: security policy changes agent behavior. Small teams should run the same task suite under nested hardening levels and separate model failures from tasks the policy made impossible.", "tags": [ "AI Agents", "Coding Agents", "Agent Evals", "Security", "Developer Workflow" ] }, { "collection": "blog", "slug": "debug-ai-reward-functions-like-incidents", "title": "Debug AI Reward Functions Like Production Incidents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/debug-ai-reward-functions-like-incidents/", "markdownUrl": "https://huecki.com/en/blog/debug-ai-reward-functions-like-incidents.md", "summary": "Bad reward functions should not be treated like prompt drafts. Treat them like production incidents: preserve traces, classify the failure, patch only the implicated logic, and rerun against the same controls.", "tags": [ "AI Engineering", "AI Agents", "Evals", "Developer Workflow", "Reinforcement Learning", "Agent Harness" ] }, { "collection": "blog", "slug": "harness-engineering-field-guide", "title": "Harness Engineering Is Systems Engineering for AI Agents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/harness-engineering-field-guide/", "markdownUrl": "https://huecki.com/en/blog/harness-engineering-field-guide.md", "summary": "Harness engineering is the work of turning a probabilistic model into a controlled system. This field guide maps the twelve engineering surfaces around the model and shows how to evaluate and evolve them.", "tags": [ "Harness Engineering", "AI Agents", "Agent Evals", "AI Security", "AI Engineering" ] }, { "collection": "blog", "slug": "ki-code-nicht-am-diff-bewerten", "title": "Bewerte KI-Code nicht am Diff", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-code-nicht-am-diff-bewerten/", "markdownUrl": "https://huecki.com/blog/ki-code-nicht-am-diff-bewerten.md", "summary": "Besseres KI-Coding entsteht nicht primär durch bessere Prompts, sondern durch den Harness um das Modell: explizite Contracts, getrennte Builder- und Reviewer-Rollen, Belege und eine Schleife, die Fehler in bessere Spezifikationen zurückführt.", "tags": [ "AI Engineering", "KI-Agenten", "Coding Agents", "Developer Workflow", "Agent Harness", "Evals" ] }, { "collection": "blog", "slug": "llm-native-developer-operational-maturity", "title": "The LLM-native developer needs more than prompts", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/llm-native-developer-operational-maturity/", "markdownUrl": "https://huecki.com/en/blog/llm-native-developer-operational-maturity.md", "summary": "The next developer skill is not writing clever prompts. It is building the operating system around LLMs: data quality, model versioning, evals, guardrails, incident response, review UX, and repo instructions agents can actually follow.", "tags": [ "AI Engineering", "LLM", "Software Architecture", "Agents", "Developer Workflow" ] }, { "collection": "blog", "slug": "llm-native-entwickler-operational-maturity", "title": "LLM-native Entwickler brauchen mehr als gute Prompts", "language": "de", "canonicalUrl": "https://huecki.com/blog/llm-native-entwickler-operational-maturity/", "markdownUrl": "https://huecki.com/blog/llm-native-entwickler-operational-maturity.md", "summary": "Die nächste Entwicklerfähigkeit ist nicht der cleverste Prompt. Es ist das Betriebssystem um LLMs herum: Datenqualität, Model-Versioning, Evals, Guardrails, Incident Response, Review-UX und Repo-Anweisungen, denen Agents wirklich folgen können.", "tags": [ "AI Engineering", "LLM", "Softwarearchitektur", "Agents", "Developer Workflow" ] }, { "collection": "blog", "slug": "measure-agentic-setups-skills", "title": "Don’t Benchmark the Model. Benchmark the Agent System.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/measure-agentic-setups-skills/", "markdownUrl": "https://huecki.com/en/blog/measure-agentic-setups-skills.md", "summary": "Agent evals should not only ask whether the final answer looked good. A useful benchmark measures the whole agent system: skill routing, tool policy, evidence, outcomes, hard-fail safety cases, regressions, cost, and production drift.", "tags": [ "AI Agents", "Agent Harness", "Evals", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "natural-language-agent-harnesses", "title": "Agent harnesses should be specs, not hidden glue code", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/natural-language-agent-harnesses/", "markdownUrl": "https://huecki.com/en/blog/natural-language-agent-harnesses.md", "summary": "Natural-Language Agent Harnesses give a useful name to an important shift: the agent policy should be an inspectable document that a runtime executes, not invisible glue hidden inside controller code.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "stop-asking-which-coding-model-is-best", "title": "Stop Asking Which Coding Model Is Best", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-asking-which-coding-model-is-best/", "markdownUrl": "https://huecki.com/en/blog/stop-asking-which-coding-model-is-best.md", "summary": "The useful question is moving from which model is best to what your agent harness can change, measure, persist, and roll back.", "tags": [ "AI Agents", "Agent Harness", "Coding Agents", "Agent Evals", "AI Engineering" ] }, { "collection": "blog", "slug": "stop-judging-ai-code-by-the-diff", "title": "Stop Judging AI Code by the Diff", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-judging-ai-code-by-the-diff/", "markdownUrl": "https://huecki.com/en/blog/stop-judging-ai-code-by-the-diff.md", "summary": "Better AI coding is not mainly about better prompts. It is about the harness around the model: explicit contracts, separate builder and reviewer roles, evidence requirements, and a loop that turns failures into better specifications.", "tags": [ "AI Engineering", "AI Agents", "Coding Agents", "Developer Workflow", "Agent Harness", "Evals" ] }, { "collection": "blog", "slug": "your-agent-harness-needs-a-release-process", "title": "Your Agent Harness Needs a Release Process", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/your-agent-harness-needs-a-release-process/", "markdownUrl": "https://huecki.com/en/blog/your-agent-harness-needs-a-release-process.md", "summary": "A practical field note on operating agent-harness changes like product releases: start from a trace-backed failure, change one bounded component, evaluate repeated trials and private holdouts, then promote through review with a rollback path.", "tags": [ "AI Agents", "Agent Harnesses", "Agent Evals", "Observability", "AI Governance" ] }, { "collection": "blog", "slug": "your-onboarding-is-why-your-team-is-vibe-coding", "title": "Your Onboarding Is Why Your Team Is Vibe Coding", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/your-onboarding-is-why-your-team-is-vibe-coding/", "markdownUrl": "https://huecki.com/en/blog/your-onboarding-is-why-your-team-is-vibe-coding.md", "summary": "Teams do not usually start vibe coding because developers became careless. They start because onboarding is broken: docs are stale, harnesses are undocumented, system knowledge lives in people’s heads, and AI turns missing context into plausible code and Markdown.", "tags": [ "AI Engineering", "Developer Onboarding", "Vibe Coding", "Developer Experience", "Internal Tools", "Agent Harness" ] }, { "collection": "tools", "slug": "agent-save-file-starter", "title": "Agent Save File Starter", "language": "de", "canonicalUrl": "https://huecki.com/tools/agent-save-file-starter/", "markdownUrl": "https://huecki.com/tools/agent-save-file-starter.md", "summary": "Ein leichtgewichtiges Projektgerüst für Agent-Kontextdateien. Es trennt harte Regeln, sichere Commands, gültige Entscheidungen und alte Hinweise, damit Agents nicht aus Chat-Müll arbeiten.", "tags": [ "Coding Agents", "Agent Memory", "Context Engineering", "AGENTS.md" ] }, { "collection": "tools", "slug": "agent-save-file-starter-en", "title": "Agent Save File Starter", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/agent-save-file-starter-en/", "markdownUrl": "https://huecki.com/en/tools/agent-save-file-starter-en.md", "summary": "A lightweight project structure for agent context files. It separates hard rules, safe commands, active decisions, and archived notes so agents do not work from chat sludge.", "tags": [ "Coding Agents", "Agent Memory", "Context Engineering", "AGENTS.md" ] }, { "collection": "tools", "slug": "is-this-agentic-change-good", "title": "Is This Agentic Change Good?", "language": "de", "canonicalUrl": "https://huecki.com/tools/is-this-agentic-change-good/", "markdownUrl": "https://huecki.com/tools/is-this-agentic-change-good.md", "summary": "Dieser Skill prüft Branches, die zukünftiges Agent-Verhalten verändern. Er sammelt lokale Git-Evidenz, bewertet Skill-Kollisionen, Kontextbudget, Sicherheitsrisiken und Runtime-Sichtbarkeit, bevor ein Merge empfohlen wird.", "tags": [ "Agent Skills", "Review", "AGENTS.md", "Safety" ] }, { "collection": "tools", "slug": "is-this-agentic-change-good-en", "title": "Is This Agentic Change Good?", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/is-this-agentic-change-good-en/", "markdownUrl": "https://huecki.com/en/tools/is-this-agentic-change-good-en.md", "summary": "This skill reviews branches that change future agent behavior. It collects local git evidence and judges skill collisions, context budget, safety risks, and runtime visibility before recommending a merge.", "tags": [ "Agent Skills", "Review", "AGENTS.md", "Safety" ] } ] }, { "id": "agent-security", "title": "Agent Security", "titleDe": "Agent Security", "description": "Secure agent runtimes, tool gates, prompt-injection controls, and auditable permissions.", "descriptionDe": "Sichere Agenten-Runtimes, Tool-Gates, Prompt-Injection-Kontrollen und auditierbare Berechtigungen.", "urls": { "en": "https://huecki.com/en/topics/agent-security/", "de": "https://huecki.com/topics/agent-security/" }, "entries": [ { "collection": "blog", "slug": "agent-plugins-npm-fuer-agent-verhalten", "title": "Agent Plugins sind npm für Agent-Verhalten — aber ohne Lockfile", "language": "de", "canonicalUrl": "https://huecki.com/blog/agent-plugins-npm-fuer-agent-verhalten/", "markdownUrl": "https://huecki.com/blog/agent-plugins-npm-fuer-agent-verhalten.md", "summary": "Agent Plugins machen aus Skills und MCP-Konfigurationen ein portables Paket. Der Guide zeigt den kleinsten Aufbau, die Integration in mehrere Clients sowie die fehlenden Produktionskontrollen für Permissions, Updates, Evals und Rollback.", "tags": [ "AI Agents", "Agent Plugins", "Agent Skills", "MCP", "Codex", "Cursor", "VS Code", "AI Security" ] }, { "collection": "blog", "slug": "agent-skills-qualitaets-gate", "title": "Agent Skills sind kein Markdown. Sie brauchen ein Qualitäts-Gate.", "language": "de", "canonicalUrl": "https://huecki.com/blog/agent-skills-qualitaets-gate/", "markdownUrl": "https://huecki.com/blog/agent-skills-qualitaets-gate.md", "summary": "Ein praktisches Qualitäts-Gate für Agent Skills: Nutzen gegen eine No-Skill-Baseline messen, Aktivierung und Trajektorie prüfen, Rechte außerhalb des Skill-Texts binden und nur versionierte, überprüfbare Kandidaten promoten.", "tags": [ "AI Agents", "Agent Skills", "Agent Evals", "AI Security", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-dont-need-longer-prompts-they-need-harnesses", "title": "Agents Don’t Need Longer Prompts. They Need Harnesses.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses/", "markdownUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses.md", "summary": "The arXiv survey Code as Agent Harness names the next shift in agent engineering: code is not only what agents generate. It is becoming the executable, inspectable, stateful runtime that makes agents reliable.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Coding Agents", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-coding-agent-harness", "title": "AGENTS.md reicht nicht: Dein Coding Agent braucht einen Harness", "language": "de", "canonicalUrl": "https://huecki.com/blog/agents-md-coding-agent-harness/", "markdownUrl": "https://huecki.com/blog/agents-md-coding-agent-harness.md", "summary": "Ein Coding Agent wird nicht durch einen magischen Prompt zuverlässig. Er braucht einen Harness: AGENTS.md, Skills, Tool-Permissions, Hooks und Evals, die merken, wenn sich sein Verhalten verschiebt.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-coding-agent-harness-en", "title": "AGENTS.md is not enough: your coding agent needs a harness", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en/", "markdownUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en.md", "summary": "A coding agent is not made reliable by one magic prompt. It needs a harness: AGENTS.md, skills, tool permissions, hooks, and evals that catch behavior drift.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "ai-agents-need-evidence-before-clicking", "title": "AI Agents Need Evidence Before They Click", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-agents-need-evidence-before-clicking/", "markdownUrl": "https://huecki.com/en/blog/ai-agents-need-evidence-before-clicking.md", "summary": "When an agent clicks, sends, pays, deletes, or extracts data, the critical truth cannot live only in model prose. Put a small evidence gate before risky tool calls: predicate, evidence type, source, decision.", "tags": [ "AI Agents", "Multimodal AI", "Browser Agents", "AI Safety", "Developer Workflow" ] }, { "collection": "blog", "slug": "ai-first-software-architecture", "title": "AI-first Architecture: Faster Decisions, Still in Control", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-first-software-architecture/", "markdownUrl": "https://huecki.com/en/blog/ai-first-software-architecture.md", "summary": "AI-first architecture does not mean the model decides. It means AI generates options, finds risks, compresses context, and the team makes a traceable decision.", "tags": [ "AI", "Software Architecture", "GEO", "Engineering" ] }, { "collection": "blog", "slug": "automatisch-generierte-agent-skills-supply-chain", "title": "Automatisch generierte Agent Skills brauchen eine Supply Chain", "language": "de", "canonicalUrl": "https://huecki.com/blog/automatisch-generierte-agent-skills-supply-chain/", "markdownUrl": "https://huecki.com/blog/automatisch-generierte-agent-skills-supply-chain.md", "summary": "Das OpenClaw-Skill-Paper ist ein starkes Signal: Agent Skills werden nicht nur manuell geschrieben, sondern aus Trajektorien, Skill-Bäumen und Transfer-Evals gelernt. Aber genau das macht eine Skill-Supply-Chain wichtiger, nicht unwichtiger.", "tags": [ "AI Engineering", "Agent Skills", "Coding Agents", "Context Engineering", "Agent Security" ] }, { "collection": "blog", "slug": "coding-agent-command-composition-risk", "title": "Your Coding Agent Can Be Tricked by Boring Shell Commands", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/coding-agent-command-composition-risk/", "markdownUrl": "https://huecki.com/en/blog/coding-agent-command-composition-risk.md", "summary": "The MOSAIC paper shifts the coding-agent security question from hostile prompts to command traces. The practical move is to audit producer-consumer state across shell commands before generated state crosses into privileged work.", "tags": [ "AI Agents", "Coding Agents", "Security", "Developer Workflow", "Failure Mode" ] }, { "collection": "blog", "slug": "coding-agents-need-hardened-harness-evals", "title": "Coding Agents Need Hardened Harness Evals", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/coding-agents-need-hardened-harness-evals/", "markdownUrl": "https://huecki.com/en/blog/coding-agents-need-hardened-harness-evals.md", "summary": "Permissive coding-agent benchmarks hide a boring production truth: security policy changes agent behavior. Small teams should run the same task suite under nested hardening levels and separate model failures from tasks the policy made impossible.", "tags": [ "AI Agents", "Coding Agents", "Agent Evals", "Security", "Developer Workflow" ] }, { "collection": "blog", "slug": "harness-engineering-field-guide", "title": "Harness Engineering Is Systems Engineering for AI Agents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/harness-engineering-field-guide/", "markdownUrl": "https://huecki.com/en/blog/harness-engineering-field-guide.md", "summary": "Harness engineering is the work of turning a probabilistic model into a controlled system. This field guide maps the twelve engineering surfaces around the model and shows how to evaluate and evolve them.", "tags": [ "Harness Engineering", "AI Agents", "Agent Evals", "AI Security", "AI Engineering" ] }, { "collection": "blog", "slug": "ki-agenten-belege-vor-dem-klick", "title": "KI-Agenten brauchen Belege, bevor sie klicken", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-agenten-belege-vor-dem-klick/", "markdownUrl": "https://huecki.com/blog/ki-agenten-belege-vor-dem-klick.md", "summary": "Wenn ein Agent klickt, sendet, kauft oder Daten extrahiert, darf die entscheidende Wahrheit nicht nur aus Modell-Prosa kommen. Baue vor riskanten Tool Calls ein kleines Evidenz-Gate: Predicate, Belegtyp, Quelle, Entscheidung.", "tags": [ "KI-Agenten", "Multimodal AI", "Browser Agents", "AI Safety", "Developer Workflow" ] }, { "collection": "blog", "slug": "ki-selbstcheck-findet-immer-etwas", "title": "Hör auf, KI zum kritischen Selbstcheck zu bitten", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-selbstcheck-findet-immer-etwas/", "markdownUrl": "https://huecki.com/blog/ki-selbstcheck-findet-immer-etwas.md", "summary": "Offene Anweisungen wie „prüf das kritisch“ belohnen das Modell ungewollt dafür, Kritik zu produzieren. Die Lösung ist nicht weniger Review, sondern kalibriertes Review: klare Kriterien, PASS_NO_CHANGE, Evidenz pro Finding, Severity-Schwellen und ein kleines Änderungsbudget.", "tags": [ "KI-Agenten", "Prompt Engineering", "Evals", "KI-Sicherheit", "Developer Workflow" ] }, { "collection": "blog", "slug": "local-llm-agent-runtime-audit", "title": "Audit Local LLM Agents Like Runtimes", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/local-llm-agent-runtime-audit/", "markdownUrl": "https://huecki.com/en/blog/local-llm-agent-runtime-audit.md", "summary": "Local LLM agents can touch shells, files, browsers, credentials, memory, and messaging tools. Treat their runtime layer as source code worth auditing, then turn static findings into a manual review queue instead of automatic verdicts.", "tags": [ "AI Security", "Local LLMs", "Agents", "Developer Workflow", "Runtime Security" ] }, { "collection": "blog", "slug": "next-prompt-is-a-workflow", "title": "The Next Prompt Is Not a Prompt. It’s a Workflow.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/next-prompt-is-a-workflow/", "markdownUrl": "https://huecki.com/en/blog/next-prompt-is-a-workflow.md", "summary": "Dynamic workflows move agent work from one chat prompt into inspectable orchestration: phases, subagents, evidence, budget, permissions, adversarial review, and stop conditions. The point is not more agents. The point is better control.", "tags": [ "AI Agents", "Claude Code", "Agent Workflows", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "should-ai-agent-remember-what-it-learned", "title": "Your AI Agent Learned Something. Should It Be Allowed to Remember It?", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/should-ai-agent-remember-what-it-learned/", "markdownUrl": "https://huecki.com/en/blog/should-ai-agent-remember-what-it-learned.md", "summary": "An agent that writes a lesson into memory, a skill, a prompt, or its own code is deploying behavior into future runs. This guide shows how to put persistent changes through evidence, eval, approval, expiry, and rollback gates.", "tags": [ "AI Agents", "Agent Memory", "Self-Improvement", "Agent Evals", "AI Security" ] }, { "collection": "tools", "slug": "is-this-agentic-change-good", "title": "Is This Agentic Change Good?", "language": "de", "canonicalUrl": "https://huecki.com/tools/is-this-agentic-change-good/", "markdownUrl": "https://huecki.com/tools/is-this-agentic-change-good.md", "summary": "Dieser Skill prüft Branches, die zukünftiges Agent-Verhalten verändern. Er sammelt lokale Git-Evidenz, bewertet Skill-Kollisionen, Kontextbudget, Sicherheitsrisiken und Runtime-Sichtbarkeit, bevor ein Merge empfohlen wird.", "tags": [ "Agent Skills", "Review", "AGENTS.md", "Safety" ] }, { "collection": "tools", "slug": "is-this-agentic-change-good-en", "title": "Is This Agentic Change Good?", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/is-this-agentic-change-good-en/", "markdownUrl": "https://huecki.com/en/tools/is-this-agentic-change-good-en.md", "summary": "This skill reviews branches that change future agent behavior. It collects local git evidence and judges skill collisions, context budget, safety risks, and runtime visibility before recommending a merge.", "tags": [ "Agent Skills", "Review", "AGENTS.md", "Safety" ] } ] }, { "id": "context-engineering", "title": "Context Engineering", "titleDe": "Context Engineering", "description": "Context, sources, schemas, skills, and task decomposition instead of longer prompts.", "descriptionDe": "Kontext, Quellen, Schemas, Skills und Aufgabenzerlegung statt nur laengere Prompts.", "urls": { "en": "https://huecki.com/en/topics/context-engineering/", "de": "https://huecki.com/topics/context-engineering/" }, "entries": [ { "collection": "blog", "slug": "agent-communication-protocols-layered-stack", "title": "Agent Protocols Are Becoming a Stack, Not a Winner-Takes-All Standard", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-communication-protocols-layered-stack/", "markdownUrl": "https://huecki.com/en/blog/agent-communication-protocols-layered-stack.md", "summary": "The useful question is not whether MCP, A2A, ACP, agents.json, Agora, ANP, LMOS, or AGNTCY wins. The useful question is which communication boundary you are designing: discovery, tool execution, task delegation, identity, transport, or runtime negotiation.", "tags": [ "AI Agents", "Agent Protocols", "MCP", "A2A", "Agent Architecture" ] }, { "collection": "blog", "slug": "agent-eval-too-short-trajectory", "title": "Your Agent Eval Is Too Short", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-eval-too-short-trajectory/", "markdownUrl": "https://huecki.com/en/blog/agent-eval-too-short-trajectory.md", "summary": "A final pass/fail score hides the part of agent work that matters most: where the run started drifting, whether it noticed, and whether it recovered. The practical replacement is a trajectory eval with checkpoints, failure labels, and recovery metrics.", "tags": [ "AI Agents", "Evals", "Developer Workflow", "Agent Harness", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-harness-is-a-binary", "title": "Your Agent's Harness Is a Binary Now", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-harness-is-a-binary/", "markdownUrl": "https://huecki.com/en/blog/agent-harness-is-a-binary.md", "summary": "Two 2026 papers from the same research lineage quietly retire prompt engineering as a discipline. The agent's system prompt is now a binary you can version, diff, and evolve with a 200-line loop. The four metrics that actually matter are not the ones your dashboard shows.", "tags": [ "AI Agents", "Agent Harness", "Evals", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-harness-needs-a-behavior-map", "title": "Your Agent Harness Needs a Behavior Map", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-harness-needs-a-behavior-map/", "markdownUrl": "https://huecki.com/en/blog/agent-harness-needs-a-behavior-map.md", "summary": "Harness Handbook points at a practical bottleneck in agent engineering: the behavior you want to change is scattered across prompts, state managers, tool calls, policy code, and tests. Build a behavior map before editing the harness.", "tags": [ "AI Agents", "Agent Harness", "Coding Agents", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-is-not-the-product-fitness-function-is", "title": "Better AI Products Need Systems, Not One Agent", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-is-not-the-product-fitness-function-is/", "markdownUrl": "https://huecki.com/en/blog/agent-is-not-the-product-fitness-function-is.md", "summary": "Better AI products come from improvement systems around the agent. This guide shows how to build one with deterministic checks, narrow scoring rubrics, private holdouts, calibrated judges, and promotion gates.", "tags": [ "AI Agents", "Agent Evals", "Self-Improvement", "Coding Agents", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-memory-tests-measure-wrong-thing", "title": "Your Agent Memory Test Is Probably Measuring the Wrong Thing", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-memory-tests-measure-wrong-thing/", "markdownUrl": "https://huecki.com/en/blog/agent-memory-tests-measure-wrong-thing.md", "summary": "Most memory evals ask whether the agent got the final answer right. MemTrace suggests a sharper unit: one durable user fact tested across age, current state, earlier state, trajectory, and contradictory evidence. That turns memory from a vague feature into a small regression suite.", "tags": [ "AI Agents", "Memory", "Evals", "RAG", "Developer Workflow" ] }, { "collection": "blog", "slug": "agent-operating-contract-not-bigger-prompt", "title": "Your Agent Needs an Operating Contract, Not a Bigger Prompt", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-operating-contract-not-bigger-prompt/", "markdownUrl": "https://huecki.com/en/blog/agent-operating-contract-not-bigger-prompt.md", "summary": "The serious agent pattern is no longer bigger prompts and more encouragement. It is an operating contract: measurable goal, bounded tools, context sources, verifier evidence, review notes, rollback path, and a skill update when the run teaches you something.", "tags": [ "AI Agents", "Coding Agents", "Agent Workflow", "Verification", "Agent Skills", "Developer Workflow" ] }, { "collection": "blog", "slug": "agent-plugins-npm-fuer-agent-verhalten", "title": "Agent Plugins sind npm für Agent-Verhalten — aber ohne Lockfile", "language": "de", "canonicalUrl": "https://huecki.com/blog/agent-plugins-npm-fuer-agent-verhalten/", "markdownUrl": "https://huecki.com/blog/agent-plugins-npm-fuer-agent-verhalten.md", "summary": "Agent Plugins machen aus Skills und MCP-Konfigurationen ein portables Paket. Der Guide zeigt den kleinsten Aufbau, die Integration in mehrere Clients sowie die fehlenden Produktionskontrollen für Permissions, Updates, Evals und Rollback.", "tags": [ "AI Agents", "Agent Plugins", "Agent Skills", "MCP", "Codex", "Cursor", "VS Code", "AI Security" ] }, { "collection": "blog", "slug": "agent-skills-qualitaets-gate", "title": "Agent Skills sind kein Markdown. Sie brauchen ein Qualitäts-Gate.", "language": "de", "canonicalUrl": "https://huecki.com/blog/agent-skills-qualitaets-gate/", "markdownUrl": "https://huecki.com/blog/agent-skills-qualitaets-gate.md", "summary": "Ein praktisches Qualitäts-Gate für Agent Skills: Nutzen gegen eine No-Skill-Baseline messen, Aktivierung und Trajektorie prüfen, Rechte außerhalb des Skill-Texts binden und nur versionierte, überprüfbare Kandidaten promoten.", "tags": [ "AI Agents", "Agent Skills", "Agent Evals", "AI Security", "Developer Workflow" ] }, { "collection": "blog", "slug": "agent-state-machines-seatbelts", "title": "Give Your Agent Seatbelts, Not a Longer Prompt", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-state-machines-seatbelts/", "markdownUrl": "https://huecki.com/en/blog/agent-state-machines-seatbelts.md", "summary": "When an agent keeps jumping from planning to editing to testing at the wrong time, the fix is not usually another paragraph of system prompt. Put the workflow into explicit states, give each state a tiny tool policy, and make phase changes visible.", "tags": [ "AI Agents", "Coding Agents", "Agent Harness", "Developer Workflow", "AI Safety" ] }, { "collection": "blog", "slug": "agenten-brauchen-runtime-contracts", "title": "Agenten brauchen Runtime Contracts, nicht längere Prompts", "language": "de", "canonicalUrl": "https://huecki.com/blog/agenten-brauchen-runtime-contracts/", "markdownUrl": "https://huecki.com/blog/agenten-brauchen-runtime-contracts.md", "summary": "Bessere Prompts machen Agenten nicht automatisch zuverlässig. Entwickler brauchen Runtime Contracts: explizite Verträge dafür, welche Tools ein Agent nutzen darf, was er erinnern darf, wann er stoppen muss und wie seine Behauptungen geprüft werden.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Developer Workflow", "KI-Workflows", "Evals" ] }, { "collection": "blog", "slug": "agents-dont-need-longer-prompts-they-need-harnesses", "title": "Agents Don’t Need Longer Prompts. They Need Harnesses.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses/", "markdownUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses.md", "summary": "The arXiv survey Code as Agent Harness names the next shift in agent engineering: code is not only what agents generate. It is becoming the executable, inspectable, stateful runtime that makes agents reliable.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Coding Agents", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-coding-agent-harness", "title": "AGENTS.md reicht nicht: Dein Coding Agent braucht einen Harness", "language": "de", "canonicalUrl": "https://huecki.com/blog/agents-md-coding-agent-harness/", "markdownUrl": "https://huecki.com/blog/agents-md-coding-agent-harness.md", "summary": "Ein Coding Agent wird nicht durch einen magischen Prompt zuverlässig. Er braucht einen Harness: AGENTS.md, Skills, Tool-Permissions, Hooks und Evals, die merken, wenn sich sein Verhalten verschiebt.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-coding-agent-harness-en", "title": "AGENTS.md is not enough: your coding agent needs a harness", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en/", "markdownUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en.md", "summary": "A coding agent is not made reliable by one magic prompt. It needs a harness: AGENTS.md, skills, tool permissions, hooks, and evals that catch behavior drift.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-control-surface", "title": "AGENTS.md ist kein Kontext. Es ist eine Steuerfläche.", "language": "de", "canonicalUrl": "https://huecki.com/blog/agents-md-control-surface/", "markdownUrl": "https://huecki.com/blog/agents-md-control-surface.md", "summary": "Die überraschende Lektion aus AGENTS.md-Benchmarks ist nicht, dass Kontextdateien nutzlos sind. Sie verändern Agent-Verhalten, manchmal in Richtung teurerer und weniger nützlicher Arbeit. Behandle sie als Steuerfläche, nicht als Repo-Handbuch.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Context Engineering", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-control-surface-en", "title": "AGENTS.md Is Not Context. It Is a Control Surface.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-md-control-surface-en/", "markdownUrl": "https://huecki.com/en/blog/agents-md-control-surface-en.md", "summary": "The surprising lesson from AGENTS.md benchmarks is not that context files are useless. It is that they change agent behavior, sometimes into more expensive and less useful work. Treat them as a control surface, not a repo manual.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Context Engineering", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-need-exit-conditions", "title": "Agents Don’t Need ‘Keep Going’. They Need Exit Conditions.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-need-exit-conditions/", "markdownUrl": "https://huecki.com/en/blog/agents-need-exit-conditions.md", "summary": "The useful lesson behind Claude Code /goal is not that agents can run forever. It is that long-running agent work needs an explicit, observable exit condition: what proves done, what stays in scope, and when to stop blocked.", "tags": [ "AI Agents", "Agent Harness", "Developer Workflow", "AI Engineering", "Evals" ] }, { "collection": "blog", "slug": "ai-agents-need-evidence-before-clicking", "title": "AI Agents Need Evidence Before They Click", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-agents-need-evidence-before-clicking/", "markdownUrl": "https://huecki.com/en/blog/ai-agents-need-evidence-before-clicking.md", "summary": "When an agent clicks, sends, pays, deletes, or extracts data, the critical truth cannot live only in model prose. Put a small evidence gate before risky tool calls: predicate, evidence type, source, decision.", "tags": [ "AI Agents", "Multimodal AI", "Browser Agents", "AI Safety", "Developer Workflow" ] }, { "collection": "blog", "slug": "ai-built-ui-needs-a-playtester", "title": "Your AI-Built UI Needs a Playtester, Not a Screenshot Review", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-built-ui-needs-a-playtester/", "markdownUrl": "https://huecki.com/en/blog/ai-built-ui-needs-a-playtester.md", "summary": "AI-generated interfaces often look finished before they behave correctly. A GUI playtester loop uses a separate browser agent to interact with the artifact, record screenshots and action logs, turn broken flows into reproducible bug reports, and rerun the same script after repairs.", "tags": [ "AI Engineering", "AI Agents", "Webwright", "Playwright", "UI Testing", "Coding Agents", "Developer Workflow", "Evals" ] }, { "collection": "blog", "slug": "ai-first-software-architecture", "title": "AI-first Architecture: Faster Decisions, Still in Control", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-first-software-architecture/", "markdownUrl": "https://huecki.com/en/blog/ai-first-software-architecture.md", "summary": "AI-first architecture does not mean the model decides. It means AI generates options, finds risks, compresses context, and the team makes a traceable decision.", "tags": [ "AI", "Software Architecture", "GEO", "Engineering" ] }, { "collection": "blog", "slug": "ai-self-check-always-finds-something", "title": "Stop Asking AI to Critically Self-Check", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-self-check-always-finds-something/", "markdownUrl": "https://huecki.com/en/blog/ai-self-check-always-finds-something.md", "summary": "Open-ended instructions like “critically self-check this” accidentally reward the model for producing criticism. The fix is not less review. It is calibrated review: explicit criteria, PASS_NO_CHANGE, evidence per finding, severity thresholds, and a tiny change budget.", "tags": [ "AI Agents", "Prompt Engineering", "Evals", "AI Safety", "Developer Workflow" ] }, { "collection": "blog", "slug": "ai-self-reflection-defends-first-answer", "title": "Your AI Agent Is Not Reflecting. It Is Defending Its First Answer", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-self-reflection-defends-first-answer/", "markdownUrl": "https://huecki.com/en/blog/ai-self-reflection-defends-first-answer.md", "summary": "Asking one agent to reconsider its answer often produces a more confident defense of the same mistake. A bounded challenger-and-judge loop can create real alternatives, but only if disagreement, stopping, and judge bias are engineered explicitly.", "tags": [ "AI Agents", "Multi-Agent Systems", "Self-Reflection", "LLM Judges", "Agent Architecture" ] }, { "collection": "blog", "slug": "ai-slop-gate-after-tests-and-lint", "title": "Put an AI Slop Gate After Tests and Lint", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-slop-gate-after-tests-and-lint/", "markdownUrl": "https://huecki.com/en/blog/ai-slop-gate-after-tests-and-lint.md", "summary": "Tests tell you whether behavior still works. Linters tell you whether code is syntactically and stylistically acceptable. An AI-slop gate catches the residue coding agents leave behind: fake comments, swallowed errors, any-casts, duplicated helpers, TODO stubs, and dead code.", "tags": [ "AI Engineering", "Coding Agents", "Developer Workflow", "Code Quality", "Evals", "Agent Harness" ] }, { "collection": "blog", "slug": "automatisch-generierte-agent-skills-supply-chain", "title": "Automatisch generierte Agent Skills brauchen eine Supply Chain", "language": "de", "canonicalUrl": "https://huecki.com/blog/automatisch-generierte-agent-skills-supply-chain/", "markdownUrl": "https://huecki.com/blog/automatisch-generierte-agent-skills-supply-chain.md", "summary": "Das OpenClaw-Skill-Paper ist ein starkes Signal: Agent Skills werden nicht nur manuell geschrieben, sondern aus Trajektorien, Skill-Bäumen und Transfer-Evals gelernt. Aber genau das macht eine Skill-Supply-Chain wichtiger, nicht unwichtiger.", "tags": [ "AI Engineering", "Agent Skills", "Coding Agents", "Context Engineering", "Agent Security" ] }, { "collection": "blog", "slug": "client-projekt-agenten-standup", "title": "gib jedem Kundenprojekt einen kleinen Agenten", "language": "de", "canonicalUrl": "https://huecki.com/blog/client-projekt-agenten-standup/", "markdownUrl": "https://huecki.com/blog/client-projekt-agenten-standup.md", "summary": "Der nützliche Move ist nicht ein Mega-Assistent für alle Kunden. Gib jedem Kundenprojekt einen kleinen, isolierten Agenten mit eigener Erinnerung, Aufgabenliste, Preview-URL-Gewohnheit und langweiligem Daily Standup.", "tags": [ "AI Agents", "Freelancing", "Client Work", "Automation", "Workflow Design" ] }, { "collection": "blog", "slug": "client-site-domain-agents", "title": "give every client project a tiny agent", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/client-site-domain-agents/", "markdownUrl": "https://huecki.com/en/blog/client-site-domain-agents.md", "summary": "The useful move is not one mega assistant for all client work. Give each client project a small, isolated agent with its own memory, tasks, preview URL habit, and boring daily standup.", "tags": [ "AI Agents", "Freelancing", "Client Work", "Automation", "Workflow Design" ] }, { "collection": "blog", "slug": "coding-agent-command-composition-risk", "title": "Your Coding Agent Can Be Tricked by Boring Shell Commands", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/coding-agent-command-composition-risk/", "markdownUrl": "https://huecki.com/en/blog/coding-agent-command-composition-risk.md", "summary": "The MOSAIC paper shifts the coding-agent security question from hostile prompts to command traces. The practical move is to audit producer-consumer state across shell commands before generated state crosses into privileged work.", "tags": [ "AI Agents", "Coding Agents", "Security", "Developer Workflow", "Failure Mode" ] }, { "collection": "blog", "slug": "coding-agent-instruktionen-testen", "title": "Teste deine Coding-Agent-Instruktionen wie Code", "language": "de", "canonicalUrl": "https://huecki.com/blog/coding-agent-instruktionen-testen/", "markdownUrl": "https://huecki.com/blog/coding-agent-instruktionen-testen.md", "summary": "Eine gute AGENTS.md ist kein Prompt-Dokument, das man einmal schreibt. Sie ist ein kleines Betriebshandbuch fuer den Agenten. Und Betriebshandbuecher werden besser, wenn man sie gegen konkrete Fehler testet.", "tags": [ "Coding Agents", "AI Engineering", "Developer Workflow", "AGENTS.md", "Evals" ] }, { "collection": "blog", "slug": "coding-agent-save-file", "title": "Dein Coding Agent braucht eine Verfassung und ein Logbuch", "language": "de", "canonicalUrl": "https://huecki.com/blog/coding-agent-save-file/", "markdownUrl": "https://huecki.com/blog/coding-agent-save-file.md", "summary": "Coding Agents werden nicht nur durch bessere Modelle nützlich. Teams brauchen eine kleine Verfassung für Agent-Verhalten und ein kuratiertes Logbuch für Projektwissen, sonst wird Memory zu Kontext-Müll.", "tags": [ "AI Engineering", "Coding Agents", "Context Engineering", "Developer Workflow", "Agent Memory" ] }, { "collection": "blog", "slug": "coding-agents-need-hardened-harness-evals", "title": "Coding Agents Need Hardened Harness Evals", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/coding-agents-need-hardened-harness-evals/", "markdownUrl": "https://huecki.com/en/blog/coding-agents-need-hardened-harness-evals.md", "summary": "Permissive coding-agent benchmarks hide a boring production truth: security policy changes agent behavior. Small teams should run the same task suite under nested hardening levels and separate model failures from tasks the policy made impossible.", "tags": [ "AI Agents", "Coding Agents", "Agent Evals", "Security", "Developer Workflow" ] }, { "collection": "blog", "slug": "debug-ai-reward-functions-like-incidents", "title": "Debug AI Reward Functions Like Production Incidents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/debug-ai-reward-functions-like-incidents/", "markdownUrl": "https://huecki.com/en/blog/debug-ai-reward-functions-like-incidents.md", "summary": "Bad reward functions should not be treated like prompt drafts. Treat them like production incidents: preserve traces, classify the failure, patch only the implicated logic, and rerun against the same controls.", "tags": [ "AI Engineering", "AI Agents", "Evals", "Developer Workflow", "Reinforcement Learning", "Agent Harness" ] }, { "collection": "blog", "slug": "dein-agent-scheiterte-vor-dem-fehler", "title": "Dein Agent scheiterte drei Schritte vor dem Fehler", "language": "de", "canonicalUrl": "https://huecki.com/blog/dein-agent-scheiterte-vor-dem-fehler/", "markdownUrl": "https://huecki.com/blog/dein-agent-scheiterte-vor-dem-fehler.md", "summary": "Bei langen Agentenläufen ist der letzte Fehler oft nur das Symptom. Der bessere Debugging-Loop sucht den frühesten kausal verantwortlichen Schritt, formuliert eine minimale Korrektur und prüft sie in einem kontrollierten Rerun.", "tags": [ "KI-Agenten", "Agent Debugging", "Observability", "Evaluation", "Developer Workflow" ] }, { "collection": "blog", "slug": "dein-ai-agent-lernt-nichts-aus-seinen-runs", "title": "Dein AI Agent lernt nichts aus seinen Runs", "language": "de", "canonicalUrl": "https://huecki.com/blog/dein-ai-agent-lernt-nichts-aus-seinen-runs/", "markdownUrl": "https://huecki.com/blog/dein-ai-agent-lernt-nichts-aus-seinen-runs.md", "summary": "Ein Agent verbessert sich nicht, nur weil seine Runs gespeichert werden. Ein Knowledge Flywheel extrahiert aus mehreren Runs belegte Lessons, prüft Widersprüche, versioniert das Ergebnis und liefert dem nächsten Agenten nur das Wissen, das zu seinem Task passt.", "tags": [ "AI Agents", "Knowledge Flywheel", "Agent Memory", "Agent Skills", "Self-Improving Agents", "Agent Evals" ] }, { "collection": "blog", "slug": "graph-engineering-routing-wird-produktverhalten", "title": "Graph Engineering beginnt dort, wo Routing zum Produktverhalten wird", "language": "de", "canonicalUrl": "https://huecki.com/blog/graph-engineering-routing-wird-produktverhalten/", "markdownUrl": "https://huecki.com/blog/graph-engineering-routing-wird-produktverhalten.md", "summary": "Graph Engineering ist nicht das Zeichnen komplexer Agentendiagramme. Es beginnt dann, wenn Retry, Eskalation, Evidenzprüfung oder menschliche Freigabe sichtbares und testbares Produktverhalten werden.", "tags": [ "Graph Engineering", "LangGraph", "AI Agents", "Agent Workflows", "Human-in-the-Loop", "AI Engineering" ] }, { "collection": "blog", "slug": "harness-engineering-field-guide", "title": "Harness Engineering Is Systems Engineering for AI Agents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/harness-engineering-field-guide/", "markdownUrl": "https://huecki.com/en/blog/harness-engineering-field-guide.md", "summary": "Harness engineering is the work of turning a probabilistic model into a controlled system. This field guide maps the twelve engineering surfaces around the model and shows how to evaluate and evolve them.", "tags": [ "Harness Engineering", "AI Agents", "Agent Evals", "AI Security", "AI Engineering" ] }, { "collection": "blog", "slug": "hermes-selbstverbessernder-agent", "title": "Hermes Agent: Self-Review statt One-Shot", "language": "de", "canonicalUrl": "https://huecki.com/blog/hermes-selbstverbessernder-agent/", "markdownUrl": "https://huecki.com/blog/hermes-selbstverbessernder-agent.md", "summary": "Hermes wird interessant, wenn ein Agent nicht nur liefert, sondern die eigene Arbeit reviewed: ausführen, messen, kritisieren, Skill umbauen, nochmal laufen lassen. Der Nutzen entsteht vor allem bei wiederholbaren Workflows.", "tags": [ "Hermes Agent", "KI-Agenten", "Self-Improvement", "Nous Research" ] }, { "collection": "blog", "slug": "hermes-self-improving-agent", "title": "Hermes Agent: Self-Review Instead of One-Shot Output", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/hermes-self-improving-agent/", "markdownUrl": "https://huecki.com/en/blog/hermes-self-improving-agent.md", "summary": "Hermes gets interesting when an agent does not only produce output, but reviews the run: execute, measure, critique, rewrite the skill, and test again. The loop pays off mainly for repeatable workflows.", "tags": [ "Hermes Agent", "AI Agents", "Self-Improvement", "Nous Research" ] }, { "collection": "blog", "slug": "ki-agenten-belege-vor-dem-klick", "title": "KI-Agenten brauchen Belege, bevor sie klicken", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-agenten-belege-vor-dem-klick/", "markdownUrl": "https://huecki.com/blog/ki-agenten-belege-vor-dem-klick.md", "summary": "Wenn ein Agent klickt, sendet, kauft oder Daten extrahiert, darf die entscheidende Wahrheit nicht nur aus Modell-Prosa kommen. Baue vor riskanten Tool Calls ein kleines Evidenz-Gate: Predicate, Belegtyp, Quelle, Entscheidung.", "tags": [ "KI-Agenten", "Multimodal AI", "Browser Agents", "AI Safety", "Developer Workflow" ] }, { "collection": "blog", "slug": "ki-code-nicht-am-diff-bewerten", "title": "Bewerte KI-Code nicht am Diff", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-code-nicht-am-diff-bewerten/", "markdownUrl": "https://huecki.com/blog/ki-code-nicht-am-diff-bewerten.md", "summary": "Besseres KI-Coding entsteht nicht primär durch bessere Prompts, sondern durch den Harness um das Modell: explizite Contracts, getrennte Builder- und Reviewer-Rollen, Belege und eine Schleife, die Fehler in bessere Spezifikationen zurückführt.", "tags": [ "AI Engineering", "KI-Agenten", "Coding Agents", "Developer Workflow", "Agent Harness", "Evals" ] }, { "collection": "blog", "slug": "ki-first-softwarearchitektur", "title": "AI-first Architektur: schneller entscheiden, sauber bleiben", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-first-softwarearchitektur/", "markdownUrl": "https://huecki.com/blog/ki-first-softwarearchitektur.md", "summary": "AI-first Architektur heißt nicht: Modell entscheidet. Es heißt: KI erzeugt Optionen, findet Risiken, verdichtet Kontext — das Team entscheidet und dokumentiert nachvollziehbar.", "tags": [ "KI", "Softwarearchitektur", "GEO", "Engineering" ] }, { "collection": "blog", "slug": "ki-generierte-ui-braucht-playtester", "title": "Deine KI-generierte UI braucht einen Playtester, keinen Screenshot-Review", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-generierte-ui-braucht-playtester/", "markdownUrl": "https://huecki.com/blog/ki-generierte-ui-braucht-playtester.md", "summary": "KI-generierte Interfaces sehen oft fertig aus, bevor sie sich korrekt verhalten. Eine GUI-Playtester-Loop schickt einen separaten Browser-Agenten in die App, protokolliert Interaktionen, speichert Screenshots und Logs, macht aus kaputten Flows reproduzierbare Bug Reports und rerunnt denselben Test nach dem Fix.", "tags": [ "AI Engineering", "KI-Agenten", "Webwright", "Playwright", "UI Testing", "Coding Agents", "Developer Workflow", "Evals" ] }, { "collection": "blog", "slug": "ki-selbstcheck-findet-immer-etwas", "title": "Hör auf, KI zum kritischen Selbstcheck zu bitten", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-selbstcheck-findet-immer-etwas/", "markdownUrl": "https://huecki.com/blog/ki-selbstcheck-findet-immer-etwas.md", "summary": "Offene Anweisungen wie „prüf das kritisch“ belohnen das Modell ungewollt dafür, Kritik zu produzieren. Die Lösung ist nicht weniger Review, sondern kalibriertes Review: klare Kriterien, PASS_NO_CHANGE, Evidenz pro Finding, Severity-Schwellen und ein kleines Änderungsbudget.", "tags": [ "KI-Agenten", "Prompt Engineering", "Evals", "KI-Sicherheit", "Developer Workflow" ] }, { "collection": "blog", "slug": "llm-native-developer-operational-maturity", "title": "The LLM-native developer needs more than prompts", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/llm-native-developer-operational-maturity/", "markdownUrl": "https://huecki.com/en/blog/llm-native-developer-operational-maturity.md", "summary": "The next developer skill is not writing clever prompts. It is building the operating system around LLMs: data quality, model versioning, evals, guardrails, incident response, review UX, and repo instructions agents can actually follow.", "tags": [ "AI Engineering", "LLM", "Software Architecture", "Agents", "Developer Workflow" ] }, { "collection": "blog", "slug": "llm-native-entwickler-operational-maturity", "title": "LLM-native Entwickler brauchen mehr als gute Prompts", "language": "de", "canonicalUrl": "https://huecki.com/blog/llm-native-entwickler-operational-maturity/", "markdownUrl": "https://huecki.com/blog/llm-native-entwickler-operational-maturity.md", "summary": "Die nächste Entwicklerfähigkeit ist nicht der cleverste Prompt. Es ist das Betriebssystem um LLMs herum: Datenqualität, Model-Versioning, Evals, Guardrails, Incident Response, Review-UX und Repo-Anweisungen, denen Agents wirklich folgen können.", "tags": [ "AI Engineering", "LLM", "Softwarearchitektur", "Agents", "Developer Workflow" ] }, { "collection": "blog", "slug": "local-llm-agent-runtime-audit", "title": "Audit Local LLM Agents Like Runtimes", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/local-llm-agent-runtime-audit/", "markdownUrl": "https://huecki.com/en/blog/local-llm-agent-runtime-audit.md", "summary": "Local LLM agents can touch shells, files, browsers, credentials, memory, and messaging tools. Treat their runtime layer as source code worth auditing, then turn static findings into a manual review queue instead of automatic verdicts.", "tags": [ "AI Security", "Local LLMs", "Agents", "Developer Workflow", "Runtime Security" ] }, { "collection": "blog", "slug": "measure-agentic-setups-skills", "title": "Don’t Benchmark the Model. Benchmark the Agent System.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/measure-agentic-setups-skills/", "markdownUrl": "https://huecki.com/en/blog/measure-agentic-setups-skills.md", "summary": "Agent evals should not only ask whether the final answer looked good. A useful benchmark measures the whole agent system: skill routing, tool policy, evidence, outcomes, hard-fail safety cases, regressions, cost, and production drift.", "tags": [ "AI Agents", "Agent Harness", "Evals", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "messenger-domain-agents-ai-governance", "title": "Wie ich aus einer Messenger-App ein Betriebssystem für Domain Agents gebaut habe", "language": "de", "canonicalUrl": "https://huecki.com/blog/messenger-domain-agents-ai-governance/", "markdownUrl": "https://huecki.com/blog/messenger-domain-agents-ai-governance.md", "summary": "Ich betreibe nicht für jede Aufgabe einen eigenen Bot. Mehrere getrennte Chat-Sessions greifen auf dieselbe agentische Runtime zu. Im zentralen Boba-DM entwerfe ich nach einem festen Factory-Playbook neue Domain-Profile aus Regeln, Memory, Skills, Tools und überprüfbaren Flows.", "tags": [ "AI Agents", "AI Governance", "Agent Workflows", "Agent Skills", "Automation", "Context Engineering" ] }, { "collection": "blog", "slug": "natural-language-agent-harnesses", "title": "Agent harnesses should be specs, not hidden glue code", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/natural-language-agent-harnesses/", "markdownUrl": "https://huecki.com/en/blog/natural-language-agent-harnesses.md", "summary": "Natural-Language Agent Harnesses give a useful name to an important shift: the agent policy should be an inspectable document that a runtime executes, not invisible glue hidden inside controller code.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "next-prompt-is-a-workflow", "title": "The Next Prompt Is Not a Prompt. It’s a Workflow.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/next-prompt-is-a-workflow/", "markdownUrl": "https://huecki.com/en/blog/next-prompt-is-a-workflow.md", "summary": "Dynamic workflows move agent work from one chat prompt into inspectable orchestration: phases, subagents, evidence, budget, permissions, adversarial review, and stop conditions. The point is not more agents. The point is better control.", "tags": [ "AI Agents", "Claude Code", "Agent Workflows", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "perfect-automated-ai-agent-eval-stack", "title": "The Perfect Automated AI Eval Stack Does Not Exist", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/perfect-automated-ai-agent-eval-stack/", "markdownUrl": "https://huecki.com/en/blog/perfect-automated-ai-agent-eval-stack.md", "summary": "The reliable eval system is not one automated judge. It is a closed loop that combines portable traces, deterministic invariants, narrow semantic judges, versioned production failures, adversarial tests, and human calibration.", "tags": [ "AI Agents", "Evals", "LLM Observability", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "prompting-2026-context-engineering", "title": "Prompting ist tot. Context zählt.", "language": "de", "canonicalUrl": "https://huecki.com/blog/prompting-2026-context-engineering/", "markdownUrl": "https://huecki.com/blog/prompting-2026-context-engineering.md", "summary": "2026 geht es nicht mehr um den einen magischen Prompt. Der bessere Ansatz: Kontext wählen, Tools und Schemas definieren, Agent-Regeln setzen und mit Evals prüfen.", "tags": [ "Prompt Engineering", "Context Engineering", "AI Agents", "KI-Workflows" ] }, { "collection": "blog", "slug": "prompting-2026-context-engineering-en", "title": "Prompting Is Dead. Context Wins.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/prompting-2026-context-engineering-en/", "markdownUrl": "https://huecki.com/en/blog/prompting-2026-context-engineering-en.md", "summary": "In 2026, good prompting is not about one magic sentence. The better approach is to curate context, define tools and schemas, set agent rules, and verify behavior with evals.", "tags": [ "Prompt Engineering", "Context Engineering", "AI Agents", "AI Workflows" ] }, { "collection": "blog", "slug": "prompting-2026-decomposition-skills-evals", "title": "Prompt Decomposition: So zerlegst du KI-Aufgaben richtig", "language": "de", "canonicalUrl": "https://huecki.com/blog/prompting-2026-decomposition-skills-evals/", "markdownUrl": "https://huecki.com/blog/prompting-2026-decomposition-skills-evals.md", "summary": "Nach Context Engineering kommt Decomposition: Entwickler sollten nicht alles in einen Prompt stopfen, sondern Aufgaben in direkte Prompts, Subtasks, Pipelines, Agent-Loops oder Skills zerlegen.", "tags": [ "Prompt Engineering", "Decomposition", "AI Agents", "Skills", "Developer Workflow" ] }, { "collection": "blog", "slug": "prompting-2026-decomposition-skills-evals-en", "title": "Prompt Decomposition: How to Break Down AI Tasks Properly", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/prompting-2026-decomposition-skills-evals-en/", "markdownUrl": "https://huecki.com/en/blog/prompting-2026-decomposition-skills-evals-en.md", "summary": "After context engineering comes decomposition: developers should stop putting everything into one prompt and instead split tasks into direct prompts, subtasks, pipelines, agent loops, or skills.", "tags": [ "Prompt Engineering", "Decomposition", "AI Agents", "Skills", "Developer Workflow" ] }, { "collection": "blog", "slug": "rag-2026-contextual-hybrid-rag", "title": "RAG 2026: Vergiss GraphRAG als Default", "language": "de", "canonicalUrl": "https://huecki.com/blog/rag-2026-contextual-hybrid-rag/", "markdownUrl": "https://huecki.com/blog/rag-2026-contextual-hybrid-rag.md", "summary": "RAG wird 2026 nicht dadurch besser, dass man GraphRAG auf alles wirft. Der robuste Default ist Contextual Hybrid RAG: saubere Ingestion, BM25 plus Embeddings, Reranking, Quellenpflicht und harte Evals.", "tags": [ "RAG", "Context Engineering", "AI Engineering", "LLM", "Evals" ] }, { "collection": "blog", "slug": "should-ai-agent-remember-what-it-learned", "title": "Your AI Agent Learned Something. Should It Be Allowed to Remember It?", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/should-ai-agent-remember-what-it-learned/", "markdownUrl": "https://huecki.com/en/blog/should-ai-agent-remember-what-it-learned.md", "summary": "An agent that writes a lesson into memory, a skill, a prompt, or its own code is deploying behavior into future runs. This guide shows how to put persistent changes through evidence, eval, approval, expiry, and rollback gates.", "tags": [ "AI Agents", "Agent Memory", "Self-Improvement", "Agent Evals", "AI Security" ] }, { "collection": "blog", "slug": "single-turn-evals-dont-teach-agent-enough", "title": "Single-Turn Evals Don’t Teach Your Agent Enough", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/single-turn-evals-dont-teach-agent-enough/", "markdownUrl": "https://huecki.com/en/blog/single-turn-evals-dont-teach-agent-enough.md", "summary": "Single-turn evals expose the first obvious skill gap, then stop teaching the system. Use multi-turn failure replay, repairable attribution, bounded edits, and a separate governance check instead.", "tags": [ "AI Agents", "Agent Skills", "Agent Evals", "AI Engineering", "Automation" ] }, { "collection": "blog", "slug": "spec-driven-context-resets-for-coding-agents", "title": "Spec-Driven Context Resets for Coding Agents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/spec-driven-context-resets-for-coding-agents/", "markdownUrl": "https://huecki.com/en/blog/spec-driven-context-resets-for-coding-agents.md", "summary": "Long agent chats rot. A better pattern is to move decisions into small spec files, clear context between layers, and let each coding-agent session read only the artifact it needs.", "tags": [ "AI Engineering", "Coding Agents", "Context Engineering", "Spec-Driven Development", "Developer Workflow" ] }, { "collection": "blog", "slug": "spec-driven-context-resets-fuer-coding-agenten", "title": "Spec-Driven Context Resets für Coding-Agenten", "language": "de", "canonicalUrl": "https://huecki.com/blog/spec-driven-context-resets-fuer-coding-agenten/", "markdownUrl": "https://huecki.com/blog/spec-driven-context-resets-fuer-coding-agenten.md", "summary": "Lange Agenten-Chats verrotten. Besser ist es, Entscheidungen in kleine Spec-Dateien zu verschieben, zwischen den Ebenen bewusst den Kontext zu resetten und jede Coding-Agent-Session nur das lesen zu lassen, was sie wirklich braucht.", "tags": [ "AI Engineering", "Coding Agents", "Context Engineering", "Spec-Driven Development", "Developer Workflow" ] }, { "collection": "blog", "slug": "sprachnachrichten-agenten-interface", "title": "Sprachnachrichten sind das beste Interface für kleine Agentenjobs", "language": "de", "canonicalUrl": "https://huecki.com/blog/sprachnachrichten-agenten-interface/", "markdownUrl": "https://huecki.com/blog/sprachnachrichten-agenten-interface.md", "summary": "Voice ist nicht gut für alles. Aber für kleine Agentenjobs ist es brutal praktisch: unterwegs eine Aufgabe diktieren, lokal transkribieren, vom bestehenden Agenten ausführen lassen und nur eine kurze Antwort zurückbekommen.", "tags": [ "AI Agents", "Voice Interface", "Automation", "Open Source", "Personal AI" ] }, { "collection": "blog", "slug": "stop-asking-which-coding-model-is-best", "title": "Stop Asking Which Coding Model Is Best", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-asking-which-coding-model-is-best/", "markdownUrl": "https://huecki.com/en/blog/stop-asking-which-coding-model-is-best.md", "summary": "The useful question is moving from which model is best to what your agent harness can change, measure, persist, and roll back.", "tags": [ "AI Agents", "Agent Harness", "Coding Agents", "Agent Evals", "AI Engineering" ] }, { "collection": "blog", "slug": "stop-judging-ai-code-by-the-diff", "title": "Stop Judging AI Code by the Diff", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-judging-ai-code-by-the-diff/", "markdownUrl": "https://huecki.com/en/blog/stop-judging-ai-code-by-the-diff.md", "summary": "Better AI coding is not mainly about better prompts. It is about the harness around the model: explicit contracts, separate builder and reviewer roles, evidence requirements, and a loop that turns failures into better specifications.", "tags": [ "AI Engineering", "AI Agents", "Coding Agents", "Developer Workflow", "Agent Harness", "Evals" ] }, { "collection": "blog", "slug": "stop-losing-agent-workflows", "title": "Stop Losing the Workflow Your Agent Just Discovered", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-losing-agent-workflows/", "markdownUrl": "https://huecki.com/en/blog/stop-losing-agent-workflows.md", "summary": "Successful agent traces should not disappear when the session ends. Extract the reusable part, wrap it in an interface, replay it, and only then admit it into a skill bank.", "tags": [ "AI Agents", "Agent Skills", "Developer Workflow", "AI Engineering", "Automation" ] }, { "collection": "blog", "slug": "stop-prompting-coding-agent-loop-spec", "title": "Stop Prompting Your Coding Agent. Give It a Loop.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-prompting-coding-agent-loop-spec/", "markdownUrl": "https://huecki.com/en/blog/stop-prompting-coding-agent-loop-spec.md", "summary": "The useful upgrade from prompt engineering is not a longer instruction block. It is a reusable loop spec: trigger, goal, allowed tools, verifier, terminal states, and memory rules. That is how repeated coding-agent work becomes operational instead of conversational.", "tags": [ "AI Agents", "Coding Agents", "Agent Skills", "Agent Memory", "Developer Workflow", "Verification" ] }, { "collection": "blog", "slug": "voice-notes-agent-interface", "title": "Voice notes are the best interface for small agent jobs", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/voice-notes-agent-interface/", "markdownUrl": "https://huecki.com/en/blog/voice-notes-agent-interface.md", "summary": "Voice is not good for everything. But for small agent jobs it is brutally useful: dictate a task while moving, transcribe it locally, let your existing agent handle it, and get only a short answer back.", "tags": [ "AI Agents", "Voice Interface", "Automation", "Open Source", "Personal AI" ] }, { "collection": "blog", "slug": "von-prompt-klappern-zu-prompt-debugging", "title": "Von 'Prompt-Klappern' zu Prompt-Debugging", "language": "de", "canonicalUrl": "https://huecki.com/blog/von-prompt-klappern-zu-prompt-debugging/", "markdownUrl": "https://huecki.com/blog/von-prompt-klappern-zu-prompt-debugging.md", "summary": "Prompt-Optimierung ist kein Suchproblem mehr — es ist ein Debugging-Problem. Contrastive Reflection liefert das Framework, um Fehler gezielt zu fangen, statt blind zu raten.", "tags": [ "AI Engineering", "Prompt Engineering", "LLM Agents", "RAG", "Evals", "KDD 2026" ] }, { "collection": "blog", "slug": "your-agent-harness-needs-a-release-process", "title": "Your Agent Harness Needs a Release Process", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/your-agent-harness-needs-a-release-process/", "markdownUrl": "https://huecki.com/en/blog/your-agent-harness-needs-a-release-process.md", "summary": "A practical field note on operating agent-harness changes like product releases: start from a trace-backed failure, change one bounded component, evaluate repeated trials and private holdouts, then promote through review with a rollback path.", "tags": [ "AI Agents", "Agent Harnesses", "Agent Evals", "Observability", "AI Governance" ] }, { "collection": "blog", "slug": "your-onboarding-is-why-your-team-is-vibe-coding", "title": "Your Onboarding Is Why Your Team Is Vibe Coding", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/your-onboarding-is-why-your-team-is-vibe-coding/", "markdownUrl": "https://huecki.com/en/blog/your-onboarding-is-why-your-team-is-vibe-coding.md", "summary": "Teams do not usually start vibe coding because developers became careless. They start because onboarding is broken: docs are stale, harnesses are undocumented, system knowledge lives in people’s heads, and AI turns missing context into plausible code and Markdown.", "tags": [ "AI Engineering", "Developer Onboarding", "Vibe Coding", "Developer Experience", "Internal Tools", "Agent Harness" ] }, { "collection": "tools", "slug": "agent-save-file-starter", "title": "Agent Save File Starter", "language": "de", "canonicalUrl": "https://huecki.com/tools/agent-save-file-starter/", "markdownUrl": "https://huecki.com/tools/agent-save-file-starter.md", "summary": "Ein leichtgewichtiges Projektgerüst für Agent-Kontextdateien. Es trennt harte Regeln, sichere Commands, gültige Entscheidungen und alte Hinweise, damit Agents nicht aus Chat-Müll arbeiten.", "tags": [ "Coding Agents", "Agent Memory", "Context Engineering", "AGENTS.md" ] }, { "collection": "tools", "slug": "agent-save-file-starter-en", "title": "Agent Save File Starter", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/agent-save-file-starter-en/", "markdownUrl": "https://huecki.com/en/tools/agent-save-file-starter-en.md", "summary": "A lightweight project structure for agent context files. It separates hard rules, safe commands, active decisions, and archived notes so agents do not work from chat sludge.", "tags": [ "Coding Agents", "Agent Memory", "Context Engineering", "AGENTS.md" ] }, { "collection": "tools", "slug": "deepresearch-skill", "title": "DeepResearch Skill", "language": "de", "canonicalUrl": "https://huecki.com/tools/deepresearch-skill/", "markdownUrl": "https://huecki.com/tools/deepresearch-skill.md", "summary": "DeepResearch zwingt Agenten aus dem schnellen Suchmaschinen-Summary-Modus heraus. Der Skill schreibt Recherche-Artefakte, trennt Primär- von Sekundärquellen und verlangt URL-gestützte Evidenz vor der Synthese.", "tags": [ "Research", "Agent Skills", "Evidence", "Sources" ] }, { "collection": "tools", "slug": "deepresearch-skill-en", "title": "DeepResearch Skill", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/deepresearch-skill-en/", "markdownUrl": "https://huecki.com/en/tools/deepresearch-skill-en.md", "summary": "DeepResearch pushes agents out of quick search-summary mode. The skill writes research artifacts, separates primary from secondary sources, and requires URL-backed evidence before synthesis.", "tags": [ "Research", "Agent Skills", "Evidence", "Sources" ] }, { "collection": "tools", "slug": "is-this-agentic-change-good", "title": "Is This Agentic Change Good?", "language": "de", "canonicalUrl": "https://huecki.com/tools/is-this-agentic-change-good/", "markdownUrl": "https://huecki.com/tools/is-this-agentic-change-good.md", "summary": "Dieser Skill prüft Branches, die zukünftiges Agent-Verhalten verändern. Er sammelt lokale Git-Evidenz, bewertet Skill-Kollisionen, Kontextbudget, Sicherheitsrisiken und Runtime-Sichtbarkeit, bevor ein Merge empfohlen wird.", "tags": [ "Agent Skills", "Review", "AGENTS.md", "Safety" ] }, { "collection": "tools", "slug": "is-this-agentic-change-good-en", "title": "Is This Agentic Change Good?", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/is-this-agentic-change-good-en/", "markdownUrl": "https://huecki.com/en/tools/is-this-agentic-change-good-en.md", "summary": "This skill reviews branches that change future agent behavior. It collects local git evidence and judges skill collisions, context budget, safety risks, and runtime visibility before recommending a merge.", "tags": [ "Agent Skills", "Review", "AGENTS.md", "Safety" ] } ] }, { "id": "agent-evals", "title": "Agent Evals", "titleDe": "Agent Evals", "description": "Measurable agent quality: benchmarks, review loops, playtests, regressions, and proof gates.", "descriptionDe": "Messbare Agentenqualitaet: Benchmarks, Review-Schleifen, Playtests, Regressionen und Proof Gates.", "urls": { "en": "https://huecki.com/en/topics/agent-evals/", "de": "https://huecki.com/topics/agent-evals/" }, "entries": [ { "collection": "blog", "slug": "agent-eval-too-short-trajectory", "title": "Your Agent Eval Is Too Short", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-eval-too-short-trajectory/", "markdownUrl": "https://huecki.com/en/blog/agent-eval-too-short-trajectory.md", "summary": "A final pass/fail score hides the part of agent work that matters most: where the run started drifting, whether it noticed, and whether it recovered. The practical replacement is a trajectory eval with checkpoints, failure labels, and recovery metrics.", "tags": [ "AI Agents", "Evals", "Developer Workflow", "Agent Harness", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-harness-is-a-binary", "title": "Your Agent's Harness Is a Binary Now", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-harness-is-a-binary/", "markdownUrl": "https://huecki.com/en/blog/agent-harness-is-a-binary.md", "summary": "Two 2026 papers from the same research lineage quietly retire prompt engineering as a discipline. The agent's system prompt is now a binary you can version, diff, and evolve with a 200-line loop. The four metrics that actually matter are not the ones your dashboard shows.", "tags": [ "AI Agents", "Agent Harness", "Evals", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-harness-needs-a-behavior-map", "title": "Your Agent Harness Needs a Behavior Map", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-harness-needs-a-behavior-map/", "markdownUrl": "https://huecki.com/en/blog/agent-harness-needs-a-behavior-map.md", "summary": "Harness Handbook points at a practical bottleneck in agent engineering: the behavior you want to change is scattered across prompts, state managers, tool calls, policy code, and tests. Build a behavior map before editing the harness.", "tags": [ "AI Agents", "Agent Harness", "Coding Agents", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-is-not-the-product-fitness-function-is", "title": "Better AI Products Need Systems, Not One Agent", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-is-not-the-product-fitness-function-is/", "markdownUrl": "https://huecki.com/en/blog/agent-is-not-the-product-fitness-function-is.md", "summary": "Better AI products come from improvement systems around the agent. This guide shows how to build one with deterministic checks, narrow scoring rubrics, private holdouts, calibrated judges, and promotion gates.", "tags": [ "AI Agents", "Agent Evals", "Self-Improvement", "Coding Agents", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-memory-tests-measure-wrong-thing", "title": "Your Agent Memory Test Is Probably Measuring the Wrong Thing", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-memory-tests-measure-wrong-thing/", "markdownUrl": "https://huecki.com/en/blog/agent-memory-tests-measure-wrong-thing.md", "summary": "Most memory evals ask whether the agent got the final answer right. MemTrace suggests a sharper unit: one durable user fact tested across age, current state, earlier state, trajectory, and contradictory evidence. That turns memory from a vague feature into a small regression suite.", "tags": [ "AI Agents", "Memory", "Evals", "RAG", "Developer Workflow" ] }, { "collection": "blog", "slug": "agent-operating-contract-not-bigger-prompt", "title": "Your Agent Needs an Operating Contract, Not a Bigger Prompt", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-operating-contract-not-bigger-prompt/", "markdownUrl": "https://huecki.com/en/blog/agent-operating-contract-not-bigger-prompt.md", "summary": "The serious agent pattern is no longer bigger prompts and more encouragement. It is an operating contract: measurable goal, bounded tools, context sources, verifier evidence, review notes, rollback path, and a skill update when the run teaches you something.", "tags": [ "AI Agents", "Coding Agents", "Agent Workflow", "Verification", "Agent Skills", "Developer Workflow" ] }, { "collection": "blog", "slug": "agent-plugins-npm-fuer-agent-verhalten", "title": "Agent Plugins sind npm für Agent-Verhalten — aber ohne Lockfile", "language": "de", "canonicalUrl": "https://huecki.com/blog/agent-plugins-npm-fuer-agent-verhalten/", "markdownUrl": "https://huecki.com/blog/agent-plugins-npm-fuer-agent-verhalten.md", "summary": "Agent Plugins machen aus Skills und MCP-Konfigurationen ein portables Paket. Der Guide zeigt den kleinsten Aufbau, die Integration in mehrere Clients sowie die fehlenden Produktionskontrollen für Permissions, Updates, Evals und Rollback.", "tags": [ "AI Agents", "Agent Plugins", "Agent Skills", "MCP", "Codex", "Cursor", "VS Code", "AI Security" ] }, { "collection": "blog", "slug": "agent-skills-qualitaets-gate", "title": "Agent Skills sind kein Markdown. Sie brauchen ein Qualitäts-Gate.", "language": "de", "canonicalUrl": "https://huecki.com/blog/agent-skills-qualitaets-gate/", "markdownUrl": "https://huecki.com/blog/agent-skills-qualitaets-gate.md", "summary": "Ein praktisches Qualitäts-Gate für Agent Skills: Nutzen gegen eine No-Skill-Baseline messen, Aktivierung und Trajektorie prüfen, Rechte außerhalb des Skill-Texts binden und nur versionierte, überprüfbare Kandidaten promoten.", "tags": [ "AI Agents", "Agent Skills", "Agent Evals", "AI Security", "Developer Workflow" ] }, { "collection": "blog", "slug": "agent-state-machines-seatbelts", "title": "Give Your Agent Seatbelts, Not a Longer Prompt", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-state-machines-seatbelts/", "markdownUrl": "https://huecki.com/en/blog/agent-state-machines-seatbelts.md", "summary": "When an agent keeps jumping from planning to editing to testing at the wrong time, the fix is not usually another paragraph of system prompt. Put the workflow into explicit states, give each state a tiny tool policy, and make phase changes visible.", "tags": [ "AI Agents", "Coding Agents", "Agent Harness", "Developer Workflow", "AI Safety" ] }, { "collection": "blog", "slug": "agenten-brauchen-runtime-contracts", "title": "Agenten brauchen Runtime Contracts, nicht längere Prompts", "language": "de", "canonicalUrl": "https://huecki.com/blog/agenten-brauchen-runtime-contracts/", "markdownUrl": "https://huecki.com/blog/agenten-brauchen-runtime-contracts.md", "summary": "Bessere Prompts machen Agenten nicht automatisch zuverlässig. Entwickler brauchen Runtime Contracts: explizite Verträge dafür, welche Tools ein Agent nutzen darf, was er erinnern darf, wann er stoppen muss und wie seine Behauptungen geprüft werden.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Developer Workflow", "KI-Workflows", "Evals" ] }, { "collection": "blog", "slug": "agents-dont-need-longer-prompts-they-need-harnesses", "title": "Agents Don’t Need Longer Prompts. They Need Harnesses.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses/", "markdownUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses.md", "summary": "The arXiv survey Code as Agent Harness names the next shift in agent engineering: code is not only what agents generate. It is becoming the executable, inspectable, stateful runtime that makes agents reliable.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Coding Agents", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-coding-agent-harness", "title": "AGENTS.md reicht nicht: Dein Coding Agent braucht einen Harness", "language": "de", "canonicalUrl": "https://huecki.com/blog/agents-md-coding-agent-harness/", "markdownUrl": "https://huecki.com/blog/agents-md-coding-agent-harness.md", "summary": "Ein Coding Agent wird nicht durch einen magischen Prompt zuverlässig. Er braucht einen Harness: AGENTS.md, Skills, Tool-Permissions, Hooks und Evals, die merken, wenn sich sein Verhalten verschiebt.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-coding-agent-harness-en", "title": "AGENTS.md is not enough: your coding agent needs a harness", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en/", "markdownUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en.md", "summary": "A coding agent is not made reliable by one magic prompt. It needs a harness: AGENTS.md, skills, tool permissions, hooks, and evals that catch behavior drift.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-control-surface", "title": "AGENTS.md ist kein Kontext. Es ist eine Steuerfläche.", "language": "de", "canonicalUrl": "https://huecki.com/blog/agents-md-control-surface/", "markdownUrl": "https://huecki.com/blog/agents-md-control-surface.md", "summary": "Die überraschende Lektion aus AGENTS.md-Benchmarks ist nicht, dass Kontextdateien nutzlos sind. Sie verändern Agent-Verhalten, manchmal in Richtung teurerer und weniger nützlicher Arbeit. Behandle sie als Steuerfläche, nicht als Repo-Handbuch.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Context Engineering", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-control-surface-en", "title": "AGENTS.md Is Not Context. It Is a Control Surface.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-md-control-surface-en/", "markdownUrl": "https://huecki.com/en/blog/agents-md-control-surface-en.md", "summary": "The surprising lesson from AGENTS.md benchmarks is not that context files are useless. It is that they change agent behavior, sometimes into more expensive and less useful work. Treat them as a control surface, not a repo manual.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Context Engineering", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-need-exit-conditions", "title": "Agents Don’t Need ‘Keep Going’. They Need Exit Conditions.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-need-exit-conditions/", "markdownUrl": "https://huecki.com/en/blog/agents-need-exit-conditions.md", "summary": "The useful lesson behind Claude Code /goal is not that agents can run forever. It is that long-running agent work needs an explicit, observable exit condition: what proves done, what stays in scope, and when to stop blocked.", "tags": [ "AI Agents", "Agent Harness", "Developer Workflow", "AI Engineering", "Evals" ] }, { "collection": "blog", "slug": "ai-built-ui-needs-a-playtester", "title": "Your AI-Built UI Needs a Playtester, Not a Screenshot Review", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-built-ui-needs-a-playtester/", "markdownUrl": "https://huecki.com/en/blog/ai-built-ui-needs-a-playtester.md", "summary": "AI-generated interfaces often look finished before they behave correctly. A GUI playtester loop uses a separate browser agent to interact with the artifact, record screenshots and action logs, turn broken flows into reproducible bug reports, and rerun the same script after repairs.", "tags": [ "AI Engineering", "AI Agents", "Webwright", "Playwright", "UI Testing", "Coding Agents", "Developer Workflow", "Evals" ] }, { "collection": "blog", "slug": "ai-first-software-architecture", "title": "AI-first Architecture: Faster Decisions, Still in Control", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-first-software-architecture/", "markdownUrl": "https://huecki.com/en/blog/ai-first-software-architecture.md", "summary": "AI-first architecture does not mean the model decides. It means AI generates options, finds risks, compresses context, and the team makes a traceable decision.", "tags": [ "AI", "Software Architecture", "GEO", "Engineering" ] }, { "collection": "blog", "slug": "ai-self-check-always-finds-something", "title": "Stop Asking AI to Critically Self-Check", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-self-check-always-finds-something/", "markdownUrl": "https://huecki.com/en/blog/ai-self-check-always-finds-something.md", "summary": "Open-ended instructions like “critically self-check this” accidentally reward the model for producing criticism. The fix is not less review. It is calibrated review: explicit criteria, PASS_NO_CHANGE, evidence per finding, severity thresholds, and a tiny change budget.", "tags": [ "AI Agents", "Prompt Engineering", "Evals", "AI Safety", "Developer Workflow" ] }, { "collection": "blog", "slug": "ai-self-reflection-defends-first-answer", "title": "Your AI Agent Is Not Reflecting. It Is Defending Its First Answer", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-self-reflection-defends-first-answer/", "markdownUrl": "https://huecki.com/en/blog/ai-self-reflection-defends-first-answer.md", "summary": "Asking one agent to reconsider its answer often produces a more confident defense of the same mistake. A bounded challenger-and-judge loop can create real alternatives, but only if disagreement, stopping, and judge bias are engineered explicitly.", "tags": [ "AI Agents", "Multi-Agent Systems", "Self-Reflection", "LLM Judges", "Agent Architecture" ] }, { "collection": "blog", "slug": "ai-slop-gate-after-tests-and-lint", "title": "Put an AI Slop Gate After Tests and Lint", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-slop-gate-after-tests-and-lint/", "markdownUrl": "https://huecki.com/en/blog/ai-slop-gate-after-tests-and-lint.md", "summary": "Tests tell you whether behavior still works. Linters tell you whether code is syntactically and stylistically acceptable. An AI-slop gate catches the residue coding agents leave behind: fake comments, swallowed errors, any-casts, duplicated helpers, TODO stubs, and dead code.", "tags": [ "AI Engineering", "Coding Agents", "Developer Workflow", "Code Quality", "Evals", "Agent Harness" ] }, { "collection": "blog", "slug": "automatisch-generierte-agent-skills-supply-chain", "title": "Automatisch generierte Agent Skills brauchen eine Supply Chain", "language": "de", "canonicalUrl": "https://huecki.com/blog/automatisch-generierte-agent-skills-supply-chain/", "markdownUrl": "https://huecki.com/blog/automatisch-generierte-agent-skills-supply-chain.md", "summary": "Das OpenClaw-Skill-Paper ist ein starkes Signal: Agent Skills werden nicht nur manuell geschrieben, sondern aus Trajektorien, Skill-Bäumen und Transfer-Evals gelernt. Aber genau das macht eine Skill-Supply-Chain wichtiger, nicht unwichtiger.", "tags": [ "AI Engineering", "Agent Skills", "Coding Agents", "Context Engineering", "Agent Security" ] }, { "collection": "blog", "slug": "client-projekt-agenten-standup", "title": "gib jedem Kundenprojekt einen kleinen Agenten", "language": "de", "canonicalUrl": "https://huecki.com/blog/client-projekt-agenten-standup/", "markdownUrl": "https://huecki.com/blog/client-projekt-agenten-standup.md", "summary": "Der nützliche Move ist nicht ein Mega-Assistent für alle Kunden. Gib jedem Kundenprojekt einen kleinen, isolierten Agenten mit eigener Erinnerung, Aufgabenliste, Preview-URL-Gewohnheit und langweiligem Daily Standup.", "tags": [ "AI Agents", "Freelancing", "Client Work", "Automation", "Workflow Design" ] }, { "collection": "blog", "slug": "client-site-domain-agents", "title": "give every client project a tiny agent", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/client-site-domain-agents/", "markdownUrl": "https://huecki.com/en/blog/client-site-domain-agents.md", "summary": "The useful move is not one mega assistant for all client work. Give each client project a small, isolated agent with its own memory, tasks, preview URL habit, and boring daily standup.", "tags": [ "AI Agents", "Freelancing", "Client Work", "Automation", "Workflow Design" ] }, { "collection": "blog", "slug": "coding-agent-instruktionen-testen", "title": "Teste deine Coding-Agent-Instruktionen wie Code", "language": "de", "canonicalUrl": "https://huecki.com/blog/coding-agent-instruktionen-testen/", "markdownUrl": "https://huecki.com/blog/coding-agent-instruktionen-testen.md", "summary": "Eine gute AGENTS.md ist kein Prompt-Dokument, das man einmal schreibt. Sie ist ein kleines Betriebshandbuch fuer den Agenten. Und Betriebshandbuecher werden besser, wenn man sie gegen konkrete Fehler testet.", "tags": [ "Coding Agents", "AI Engineering", "Developer Workflow", "AGENTS.md", "Evals" ] }, { "collection": "blog", "slug": "coding-agents-need-hardened-harness-evals", "title": "Coding Agents Need Hardened Harness Evals", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/coding-agents-need-hardened-harness-evals/", "markdownUrl": "https://huecki.com/en/blog/coding-agents-need-hardened-harness-evals.md", "summary": "Permissive coding-agent benchmarks hide a boring production truth: security policy changes agent behavior. Small teams should run the same task suite under nested hardening levels and separate model failures from tasks the policy made impossible.", "tags": [ "AI Agents", "Coding Agents", "Agent Evals", "Security", "Developer Workflow" ] }, { "collection": "blog", "slug": "debug-ai-reward-functions-like-incidents", "title": "Debug AI Reward Functions Like Production Incidents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/debug-ai-reward-functions-like-incidents/", "markdownUrl": "https://huecki.com/en/blog/debug-ai-reward-functions-like-incidents.md", "summary": "Bad reward functions should not be treated like prompt drafts. Treat them like production incidents: preserve traces, classify the failure, patch only the implicated logic, and rerun against the same controls.", "tags": [ "AI Engineering", "AI Agents", "Evals", "Developer Workflow", "Reinforcement Learning", "Agent Harness" ] }, { "collection": "blog", "slug": "dein-agent-scheiterte-vor-dem-fehler", "title": "Dein Agent scheiterte drei Schritte vor dem Fehler", "language": "de", "canonicalUrl": "https://huecki.com/blog/dein-agent-scheiterte-vor-dem-fehler/", "markdownUrl": "https://huecki.com/blog/dein-agent-scheiterte-vor-dem-fehler.md", "summary": "Bei langen Agentenläufen ist der letzte Fehler oft nur das Symptom. Der bessere Debugging-Loop sucht den frühesten kausal verantwortlichen Schritt, formuliert eine minimale Korrektur und prüft sie in einem kontrollierten Rerun.", "tags": [ "KI-Agenten", "Agent Debugging", "Observability", "Evaluation", "Developer Workflow" ] }, { "collection": "blog", "slug": "dein-ai-agent-lernt-nichts-aus-seinen-runs", "title": "Dein AI Agent lernt nichts aus seinen Runs", "language": "de", "canonicalUrl": "https://huecki.com/blog/dein-ai-agent-lernt-nichts-aus-seinen-runs/", "markdownUrl": "https://huecki.com/blog/dein-ai-agent-lernt-nichts-aus-seinen-runs.md", "summary": "Ein Agent verbessert sich nicht, nur weil seine Runs gespeichert werden. Ein Knowledge Flywheel extrahiert aus mehreren Runs belegte Lessons, prüft Widersprüche, versioniert das Ergebnis und liefert dem nächsten Agenten nur das Wissen, das zu seinem Task passt.", "tags": [ "AI Agents", "Knowledge Flywheel", "Agent Memory", "Agent Skills", "Self-Improving Agents", "Agent Evals" ] }, { "collection": "blog", "slug": "graph-engineering-routing-wird-produktverhalten", "title": "Graph Engineering beginnt dort, wo Routing zum Produktverhalten wird", "language": "de", "canonicalUrl": "https://huecki.com/blog/graph-engineering-routing-wird-produktverhalten/", "markdownUrl": "https://huecki.com/blog/graph-engineering-routing-wird-produktverhalten.md", "summary": "Graph Engineering ist nicht das Zeichnen komplexer Agentendiagramme. Es beginnt dann, wenn Retry, Eskalation, Evidenzprüfung oder menschliche Freigabe sichtbares und testbares Produktverhalten werden.", "tags": [ "Graph Engineering", "LangGraph", "AI Agents", "Agent Workflows", "Human-in-the-Loop", "AI Engineering" ] }, { "collection": "blog", "slug": "harness-engineering-field-guide", "title": "Harness Engineering Is Systems Engineering for AI Agents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/harness-engineering-field-guide/", "markdownUrl": "https://huecki.com/en/blog/harness-engineering-field-guide.md", "summary": "Harness engineering is the work of turning a probabilistic model into a controlled system. This field guide maps the twelve engineering surfaces around the model and shows how to evaluate and evolve them.", "tags": [ "Harness Engineering", "AI Agents", "Agent Evals", "AI Security", "AI Engineering" ] }, { "collection": "blog", "slug": "hermes-selbstverbessernder-agent", "title": "Hermes Agent: Self-Review statt One-Shot", "language": "de", "canonicalUrl": "https://huecki.com/blog/hermes-selbstverbessernder-agent/", "markdownUrl": "https://huecki.com/blog/hermes-selbstverbessernder-agent.md", "summary": "Hermes wird interessant, wenn ein Agent nicht nur liefert, sondern die eigene Arbeit reviewed: ausführen, messen, kritisieren, Skill umbauen, nochmal laufen lassen. Der Nutzen entsteht vor allem bei wiederholbaren Workflows.", "tags": [ "Hermes Agent", "KI-Agenten", "Self-Improvement", "Nous Research" ] }, { "collection": "blog", "slug": "hermes-self-improving-agent", "title": "Hermes Agent: Self-Review Instead of One-Shot Output", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/hermes-self-improving-agent/", "markdownUrl": "https://huecki.com/en/blog/hermes-self-improving-agent.md", "summary": "Hermes gets interesting when an agent does not only produce output, but reviews the run: execute, measure, critique, rewrite the skill, and test again. The loop pays off mainly for repeatable workflows.", "tags": [ "Hermes Agent", "AI Agents", "Self-Improvement", "Nous Research" ] }, { "collection": "blog", "slug": "ki-code-nicht-am-diff-bewerten", "title": "Bewerte KI-Code nicht am Diff", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-code-nicht-am-diff-bewerten/", "markdownUrl": "https://huecki.com/blog/ki-code-nicht-am-diff-bewerten.md", "summary": "Besseres KI-Coding entsteht nicht primär durch bessere Prompts, sondern durch den Harness um das Modell: explizite Contracts, getrennte Builder- und Reviewer-Rollen, Belege und eine Schleife, die Fehler in bessere Spezifikationen zurückführt.", "tags": [ "AI Engineering", "KI-Agenten", "Coding Agents", "Developer Workflow", "Agent Harness", "Evals" ] }, { "collection": "blog", "slug": "ki-first-softwarearchitektur", "title": "AI-first Architektur: schneller entscheiden, sauber bleiben", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-first-softwarearchitektur/", "markdownUrl": "https://huecki.com/blog/ki-first-softwarearchitektur.md", "summary": "AI-first Architektur heißt nicht: Modell entscheidet. Es heißt: KI erzeugt Optionen, findet Risiken, verdichtet Kontext — das Team entscheidet und dokumentiert nachvollziehbar.", "tags": [ "KI", "Softwarearchitektur", "GEO", "Engineering" ] }, { "collection": "blog", "slug": "ki-generierte-ui-braucht-playtester", "title": "Deine KI-generierte UI braucht einen Playtester, keinen Screenshot-Review", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-generierte-ui-braucht-playtester/", "markdownUrl": "https://huecki.com/blog/ki-generierte-ui-braucht-playtester.md", "summary": "KI-generierte Interfaces sehen oft fertig aus, bevor sie sich korrekt verhalten. Eine GUI-Playtester-Loop schickt einen separaten Browser-Agenten in die App, protokolliert Interaktionen, speichert Screenshots und Logs, macht aus kaputten Flows reproduzierbare Bug Reports und rerunnt denselben Test nach dem Fix.", "tags": [ "AI Engineering", "KI-Agenten", "Webwright", "Playwright", "UI Testing", "Coding Agents", "Developer Workflow", "Evals" ] }, { "collection": "blog", "slug": "ki-selbstcheck-findet-immer-etwas", "title": "Hör auf, KI zum kritischen Selbstcheck zu bitten", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-selbstcheck-findet-immer-etwas/", "markdownUrl": "https://huecki.com/blog/ki-selbstcheck-findet-immer-etwas.md", "summary": "Offene Anweisungen wie „prüf das kritisch“ belohnen das Modell ungewollt dafür, Kritik zu produzieren. Die Lösung ist nicht weniger Review, sondern kalibriertes Review: klare Kriterien, PASS_NO_CHANGE, Evidenz pro Finding, Severity-Schwellen und ein kleines Änderungsbudget.", "tags": [ "KI-Agenten", "Prompt Engineering", "Evals", "KI-Sicherheit", "Developer Workflow" ] }, { "collection": "blog", "slug": "llm-native-developer-operational-maturity", "title": "The LLM-native developer needs more than prompts", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/llm-native-developer-operational-maturity/", "markdownUrl": "https://huecki.com/en/blog/llm-native-developer-operational-maturity.md", "summary": "The next developer skill is not writing clever prompts. It is building the operating system around LLMs: data quality, model versioning, evals, guardrails, incident response, review UX, and repo instructions agents can actually follow.", "tags": [ "AI Engineering", "LLM", "Software Architecture", "Agents", "Developer Workflow" ] }, { "collection": "blog", "slug": "llm-native-entwickler-operational-maturity", "title": "LLM-native Entwickler brauchen mehr als gute Prompts", "language": "de", "canonicalUrl": "https://huecki.com/blog/llm-native-entwickler-operational-maturity/", "markdownUrl": "https://huecki.com/blog/llm-native-entwickler-operational-maturity.md", "summary": "Die nächste Entwicklerfähigkeit ist nicht der cleverste Prompt. Es ist das Betriebssystem um LLMs herum: Datenqualität, Model-Versioning, Evals, Guardrails, Incident Response, Review-UX und Repo-Anweisungen, denen Agents wirklich folgen können.", "tags": [ "AI Engineering", "LLM", "Softwarearchitektur", "Agents", "Developer Workflow" ] }, { "collection": "blog", "slug": "local-llm-agent-runtime-audit", "title": "Audit Local LLM Agents Like Runtimes", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/local-llm-agent-runtime-audit/", "markdownUrl": "https://huecki.com/en/blog/local-llm-agent-runtime-audit.md", "summary": "Local LLM agents can touch shells, files, browsers, credentials, memory, and messaging tools. Treat their runtime layer as source code worth auditing, then turn static findings into a manual review queue instead of automatic verdicts.", "tags": [ "AI Security", "Local LLMs", "Agents", "Developer Workflow", "Runtime Security" ] }, { "collection": "blog", "slug": "measure-agentic-setups-skills", "title": "Don’t Benchmark the Model. Benchmark the Agent System.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/measure-agentic-setups-skills/", "markdownUrl": "https://huecki.com/en/blog/measure-agentic-setups-skills.md", "summary": "Agent evals should not only ask whether the final answer looked good. A useful benchmark measures the whole agent system: skill routing, tool policy, evidence, outcomes, hard-fail safety cases, regressions, cost, and production drift.", "tags": [ "AI Agents", "Agent Harness", "Evals", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "natural-language-agent-harnesses", "title": "Agent harnesses should be specs, not hidden glue code", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/natural-language-agent-harnesses/", "markdownUrl": "https://huecki.com/en/blog/natural-language-agent-harnesses.md", "summary": "Natural-Language Agent Harnesses give a useful name to an important shift: the agent policy should be an inspectable document that a runtime executes, not invisible glue hidden inside controller code.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "next-prompt-is-a-workflow", "title": "The Next Prompt Is Not a Prompt. It’s a Workflow.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/next-prompt-is-a-workflow/", "markdownUrl": "https://huecki.com/en/blog/next-prompt-is-a-workflow.md", "summary": "Dynamic workflows move agent work from one chat prompt into inspectable orchestration: phases, subagents, evidence, budget, permissions, adversarial review, and stop conditions. The point is not more agents. The point is better control.", "tags": [ "AI Agents", "Claude Code", "Agent Workflows", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "perfect-automated-ai-agent-eval-stack", "title": "The Perfect Automated AI Eval Stack Does Not Exist", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/perfect-automated-ai-agent-eval-stack/", "markdownUrl": "https://huecki.com/en/blog/perfect-automated-ai-agent-eval-stack.md", "summary": "The reliable eval system is not one automated judge. It is a closed loop that combines portable traces, deterministic invariants, narrow semantic judges, versioned production failures, adversarial tests, and human calibration.", "tags": [ "AI Agents", "Evals", "LLM Observability", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "prompting-2026-context-engineering", "title": "Prompting ist tot. Context zählt.", "language": "de", "canonicalUrl": "https://huecki.com/blog/prompting-2026-context-engineering/", "markdownUrl": "https://huecki.com/blog/prompting-2026-context-engineering.md", "summary": "2026 geht es nicht mehr um den einen magischen Prompt. Der bessere Ansatz: Kontext wählen, Tools und Schemas definieren, Agent-Regeln setzen und mit Evals prüfen.", "tags": [ "Prompt Engineering", "Context Engineering", "AI Agents", "KI-Workflows" ] }, { "collection": "blog", "slug": "prompting-2026-context-engineering-en", "title": "Prompting Is Dead. Context Wins.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/prompting-2026-context-engineering-en/", "markdownUrl": "https://huecki.com/en/blog/prompting-2026-context-engineering-en.md", "summary": "In 2026, good prompting is not about one magic sentence. The better approach is to curate context, define tools and schemas, set agent rules, and verify behavior with evals.", "tags": [ "Prompt Engineering", "Context Engineering", "AI Agents", "AI Workflows" ] }, { "collection": "blog", "slug": "prompting-2026-decomposition-skills-evals", "title": "Prompt Decomposition: So zerlegst du KI-Aufgaben richtig", "language": "de", "canonicalUrl": "https://huecki.com/blog/prompting-2026-decomposition-skills-evals/", "markdownUrl": "https://huecki.com/blog/prompting-2026-decomposition-skills-evals.md", "summary": "Nach Context Engineering kommt Decomposition: Entwickler sollten nicht alles in einen Prompt stopfen, sondern Aufgaben in direkte Prompts, Subtasks, Pipelines, Agent-Loops oder Skills zerlegen.", "tags": [ "Prompt Engineering", "Decomposition", "AI Agents", "Skills", "Developer Workflow" ] }, { "collection": "blog", "slug": "prompting-2026-decomposition-skills-evals-en", "title": "Prompt Decomposition: How to Break Down AI Tasks Properly", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/prompting-2026-decomposition-skills-evals-en/", "markdownUrl": "https://huecki.com/en/blog/prompting-2026-decomposition-skills-evals-en.md", "summary": "After context engineering comes decomposition: developers should stop putting everything into one prompt and instead split tasks into direct prompts, subtasks, pipelines, agent loops, or skills.", "tags": [ "Prompt Engineering", "Decomposition", "AI Agents", "Skills", "Developer Workflow" ] }, { "collection": "blog", "slug": "rag-2026-contextual-hybrid-rag", "title": "RAG 2026: Vergiss GraphRAG als Default", "language": "de", "canonicalUrl": "https://huecki.com/blog/rag-2026-contextual-hybrid-rag/", "markdownUrl": "https://huecki.com/blog/rag-2026-contextual-hybrid-rag.md", "summary": "RAG wird 2026 nicht dadurch besser, dass man GraphRAG auf alles wirft. Der robuste Default ist Contextual Hybrid RAG: saubere Ingestion, BM25 plus Embeddings, Reranking, Quellenpflicht und harte Evals.", "tags": [ "RAG", "Context Engineering", "AI Engineering", "LLM", "Evals" ] }, { "collection": "blog", "slug": "should-ai-agent-remember-what-it-learned", "title": "Your AI Agent Learned Something. Should It Be Allowed to Remember It?", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/should-ai-agent-remember-what-it-learned/", "markdownUrl": "https://huecki.com/en/blog/should-ai-agent-remember-what-it-learned.md", "summary": "An agent that writes a lesson into memory, a skill, a prompt, or its own code is deploying behavior into future runs. This guide shows how to put persistent changes through evidence, eval, approval, expiry, and rollback gates.", "tags": [ "AI Agents", "Agent Memory", "Self-Improvement", "Agent Evals", "AI Security" ] }, { "collection": "blog", "slug": "simple-graph-rag-hallucinations", "title": "Simple Graph-RAG halbiert Halluzinationen – ohne komplexes Knowledge Graph Schema", "language": "de", "canonicalUrl": "https://huecki.com/blog/simple-graph-rag-hallucinations/", "markdownUrl": "https://huecki.com/blog/simple-graph-rag-hallucinations.md", "summary": "Vector-RAG + einfacher Document-Graph halbiert Halluzinationen auf komplexen QA-Aufgaben im MoNaCo-Benchmark. Die Erkenntnis: Man braucht kein komplexes Knowledge Graph Schema – nur strukturierte Navigation über Dokumente.", "tags": [ "RAG", "GraphRAG", "Hallucinations", "Knowledge Graph", "AI Engineering", "QA" ] }, { "collection": "blog", "slug": "single-turn-evals-dont-teach-agent-enough", "title": "Single-Turn Evals Don’t Teach Your Agent Enough", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/single-turn-evals-dont-teach-agent-enough/", "markdownUrl": "https://huecki.com/en/blog/single-turn-evals-dont-teach-agent-enough.md", "summary": "Single-turn evals expose the first obvious skill gap, then stop teaching the system. Use multi-turn failure replay, repairable attribution, bounded edits, and a separate governance check instead.", "tags": [ "AI Agents", "Agent Skills", "Agent Evals", "AI Engineering", "Automation" ] }, { "collection": "blog", "slug": "stop-asking-which-coding-model-is-best", "title": "Stop Asking Which Coding Model Is Best", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-asking-which-coding-model-is-best/", "markdownUrl": "https://huecki.com/en/blog/stop-asking-which-coding-model-is-best.md", "summary": "The useful question is moving from which model is best to what your agent harness can change, measure, persist, and roll back.", "tags": [ "AI Agents", "Agent Harness", "Coding Agents", "Agent Evals", "AI Engineering" ] }, { "collection": "blog", "slug": "stop-judging-ai-code-by-the-diff", "title": "Stop Judging AI Code by the Diff", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-judging-ai-code-by-the-diff/", "markdownUrl": "https://huecki.com/en/blog/stop-judging-ai-code-by-the-diff.md", "summary": "Better AI coding is not mainly about better prompts. It is about the harness around the model: explicit contracts, separate builder and reviewer roles, evidence requirements, and a loop that turns failures into better specifications.", "tags": [ "AI Engineering", "AI Agents", "Coding Agents", "Developer Workflow", "Agent Harness", "Evals" ] }, { "collection": "blog", "slug": "von-prompt-klappern-zu-prompt-debugging", "title": "Von 'Prompt-Klappern' zu Prompt-Debugging", "language": "de", "canonicalUrl": "https://huecki.com/blog/von-prompt-klappern-zu-prompt-debugging/", "markdownUrl": "https://huecki.com/blog/von-prompt-klappern-zu-prompt-debugging.md", "summary": "Prompt-Optimierung ist kein Suchproblem mehr — es ist ein Debugging-Problem. Contrastive Reflection liefert das Framework, um Fehler gezielt zu fangen, statt blind zu raten.", "tags": [ "AI Engineering", "Prompt Engineering", "LLM Agents", "RAG", "Evals", "KDD 2026" ] }, { "collection": "blog", "slug": "your-agent-harness-needs-a-release-process", "title": "Your Agent Harness Needs a Release Process", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/your-agent-harness-needs-a-release-process/", "markdownUrl": "https://huecki.com/en/blog/your-agent-harness-needs-a-release-process.md", "summary": "A practical field note on operating agent-harness changes like product releases: start from a trace-backed failure, change one bounded component, evaluate repeated trials and private holdouts, then promote through review with a rollback path.", "tags": [ "AI Agents", "Agent Harnesses", "Agent Evals", "Observability", "AI Governance" ] }, { "collection": "tools", "slug": "is-this-agentic-change-good", "title": "Is This Agentic Change Good?", "language": "de", "canonicalUrl": "https://huecki.com/tools/is-this-agentic-change-good/", "markdownUrl": "https://huecki.com/tools/is-this-agentic-change-good.md", "summary": "Dieser Skill prüft Branches, die zukünftiges Agent-Verhalten verändern. Er sammelt lokale Git-Evidenz, bewertet Skill-Kollisionen, Kontextbudget, Sicherheitsrisiken und Runtime-Sichtbarkeit, bevor ein Merge empfohlen wird.", "tags": [ "Agent Skills", "Review", "AGENTS.md", "Safety" ] }, { "collection": "tools", "slug": "is-this-agentic-change-good-en", "title": "Is This Agentic Change Good?", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/is-this-agentic-change-good-en/", "markdownUrl": "https://huecki.com/en/tools/is-this-agentic-change-good-en.md", "summary": "This skill reviews branches that change future agent behavior. It collects local git evidence and judges skill collisions, context budget, safety risks, and runtime visibility before recommending a merge.", "tags": [ "Agent Skills", "Review", "AGENTS.md", "Safety" ] } ] }, { "id": "llm-native-engineering", "title": "LLM-native Engineering", "titleDe": "LLM-native Engineering", "description": "Production-grade LLM development with evals, observability, UX trust, cost, and ownership.", "descriptionDe": "Produktionsreife LLM-Entwicklung mit Evals, Observability, UX-Vertrauen, Kosten und Ownership.", "urls": { "en": "https://huecki.com/en/topics/llm-native-engineering/", "de": "https://huecki.com/topics/llm-native-engineering/" }, "entries": [ { "collection": "blog", "slug": "agent-communication-protocols-layered-stack", "title": "Agent Protocols Are Becoming a Stack, Not a Winner-Takes-All Standard", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-communication-protocols-layered-stack/", "markdownUrl": "https://huecki.com/en/blog/agent-communication-protocols-layered-stack.md", "summary": "The useful question is not whether MCP, A2A, ACP, agents.json, Agora, ANP, LMOS, or AGNTCY wins. The useful question is which communication boundary you are designing: discovery, tool execution, task delegation, identity, transport, or runtime negotiation.", "tags": [ "AI Agents", "Agent Protocols", "MCP", "A2A", "Agent Architecture" ] }, { "collection": "blog", "slug": "agent-harness-is-a-binary", "title": "Your Agent's Harness Is a Binary Now", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-harness-is-a-binary/", "markdownUrl": "https://huecki.com/en/blog/agent-harness-is-a-binary.md", "summary": "Two 2026 papers from the same research lineage quietly retire prompt engineering as a discipline. The agent's system prompt is now a binary you can version, diff, and evolve with a 200-line loop. The four metrics that actually matter are not the ones your dashboard shows.", "tags": [ "AI Agents", "Agent Harness", "Evals", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-is-not-the-product-fitness-function-is", "title": "Better AI Products Need Systems, Not One Agent", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-is-not-the-product-fitness-function-is/", "markdownUrl": "https://huecki.com/en/blog/agent-is-not-the-product-fitness-function-is.md", "summary": "Better AI products come from improvement systems around the agent. This guide shows how to build one with deterministic checks, narrow scoring rubrics, private holdouts, calibrated judges, and promotion gates.", "tags": [ "AI Agents", "Agent Evals", "Self-Improvement", "Coding Agents", "AI Engineering" ] }, { "collection": "blog", "slug": "agent-operating-contract-not-bigger-prompt", "title": "Your Agent Needs an Operating Contract, Not a Bigger Prompt", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-operating-contract-not-bigger-prompt/", "markdownUrl": "https://huecki.com/en/blog/agent-operating-contract-not-bigger-prompt.md", "summary": "The serious agent pattern is no longer bigger prompts and more encouragement. It is an operating contract: measurable goal, bounded tools, context sources, verifier evidence, review notes, rollback path, and a skill update when the run teaches you something.", "tags": [ "AI Agents", "Coding Agents", "Agent Workflow", "Verification", "Agent Skills", "Developer Workflow" ] }, { "collection": "blog", "slug": "agenten-brauchen-runtime-contracts", "title": "Agenten brauchen Runtime Contracts, nicht längere Prompts", "language": "de", "canonicalUrl": "https://huecki.com/blog/agenten-brauchen-runtime-contracts/", "markdownUrl": "https://huecki.com/blog/agenten-brauchen-runtime-contracts.md", "summary": "Bessere Prompts machen Agenten nicht automatisch zuverlässig. Entwickler brauchen Runtime Contracts: explizite Verträge dafür, welche Tools ein Agent nutzen darf, was er erinnern darf, wann er stoppen muss und wie seine Behauptungen geprüft werden.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Developer Workflow", "KI-Workflows", "Evals" ] }, { "collection": "blog", "slug": "agents-dont-need-longer-prompts-they-need-harnesses", "title": "Agents Don’t Need Longer Prompts. They Need Harnesses.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses/", "markdownUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses.md", "summary": "The arXiv survey Code as Agent Harness names the next shift in agent engineering: code is not only what agents generate. It is becoming the executable, inspectable, stateful runtime that makes agents reliable.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Coding Agents", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-coding-agent-harness", "title": "AGENTS.md reicht nicht: Dein Coding Agent braucht einen Harness", "language": "de", "canonicalUrl": "https://huecki.com/blog/agents-md-coding-agent-harness/", "markdownUrl": "https://huecki.com/blog/agents-md-coding-agent-harness.md", "summary": "Ein Coding Agent wird nicht durch einen magischen Prompt zuverlässig. Er braucht einen Harness: AGENTS.md, Skills, Tool-Permissions, Hooks und Evals, die merken, wenn sich sein Verhalten verschiebt.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-coding-agent-harness-en", "title": "AGENTS.md is not enough: your coding agent needs a harness", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en/", "markdownUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en.md", "summary": "A coding agent is not made reliable by one magic prompt. It needs a harness: AGENTS.md, skills, tool permissions, hooks, and evals that catch behavior drift.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-control-surface", "title": "AGENTS.md ist kein Kontext. Es ist eine Steuerfläche.", "language": "de", "canonicalUrl": "https://huecki.com/blog/agents-md-control-surface/", "markdownUrl": "https://huecki.com/blog/agents-md-control-surface.md", "summary": "Die überraschende Lektion aus AGENTS.md-Benchmarks ist nicht, dass Kontextdateien nutzlos sind. Sie verändern Agent-Verhalten, manchmal in Richtung teurerer und weniger nützlicher Arbeit. Behandle sie als Steuerfläche, nicht als Repo-Handbuch.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Context Engineering", "Developer Workflow" ] }, { "collection": "blog", "slug": "agents-md-control-surface-en", "title": "AGENTS.md Is Not Context. It Is a Control Surface.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-md-control-surface-en/", "markdownUrl": "https://huecki.com/en/blog/agents-md-control-surface-en.md", "summary": "The surprising lesson from AGENTS.md benchmarks is not that context files are useless. It is that they change agent behavior, sometimes into more expensive and less useful work. Treat them as a control surface, not a repo manual.", "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Context Engineering", "Developer Workflow" ] }, { "collection": "blog", "slug": "ai-built-ui-needs-a-playtester", "title": "Your AI-Built UI Needs a Playtester, Not a Screenshot Review", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-built-ui-needs-a-playtester/", "markdownUrl": "https://huecki.com/en/blog/ai-built-ui-needs-a-playtester.md", "summary": "AI-generated interfaces often look finished before they behave correctly. A GUI playtester loop uses a separate browser agent to interact with the artifact, record screenshots and action logs, turn broken flows into reproducible bug reports, and rerun the same script after repairs.", "tags": [ "AI Engineering", "AI Agents", "Webwright", "Playwright", "UI Testing", "Coding Agents", "Developer Workflow", "Evals" ] }, { "collection": "blog", "slug": "ai-first-software-architecture", "title": "AI-first Architecture: Faster Decisions, Still in Control", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-first-software-architecture/", "markdownUrl": "https://huecki.com/en/blog/ai-first-software-architecture.md", "summary": "AI-first architecture does not mean the model decides. It means AI generates options, finds risks, compresses context, and the team makes a traceable decision.", "tags": [ "AI", "Software Architecture", "GEO", "Engineering" ] }, { "collection": "blog", "slug": "ai-self-reflection-defends-first-answer", "title": "Your AI Agent Is Not Reflecting. It Is Defending Its First Answer", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-self-reflection-defends-first-answer/", "markdownUrl": "https://huecki.com/en/blog/ai-self-reflection-defends-first-answer.md", "summary": "Asking one agent to reconsider its answer often produces a more confident defense of the same mistake. A bounded challenger-and-judge loop can create real alternatives, but only if disagreement, stopping, and judge bias are engineered explicitly.", "tags": [ "AI Agents", "Multi-Agent Systems", "Self-Reflection", "LLM Judges", "Agent Architecture" ] }, { "collection": "blog", "slug": "ai-slop-gate-after-tests-and-lint", "title": "Put an AI Slop Gate After Tests and Lint", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-slop-gate-after-tests-and-lint/", "markdownUrl": "https://huecki.com/en/blog/ai-slop-gate-after-tests-and-lint.md", "summary": "Tests tell you whether behavior still works. Linters tell you whether code is syntactically and stylistically acceptable. An AI-slop gate catches the residue coding agents leave behind: fake comments, swallowed errors, any-casts, duplicated helpers, TODO stubs, and dead code.", "tags": [ "AI Engineering", "Coding Agents", "Developer Workflow", "Code Quality", "Evals", "Agent Harness" ] }, { "collection": "blog", "slug": "automatisch-generierte-agent-skills-supply-chain", "title": "Automatisch generierte Agent Skills brauchen eine Supply Chain", "language": "de", "canonicalUrl": "https://huecki.com/blog/automatisch-generierte-agent-skills-supply-chain/", "markdownUrl": "https://huecki.com/blog/automatisch-generierte-agent-skills-supply-chain.md", "summary": "Das OpenClaw-Skill-Paper ist ein starkes Signal: Agent Skills werden nicht nur manuell geschrieben, sondern aus Trajektorien, Skill-Bäumen und Transfer-Evals gelernt. Aber genau das macht eine Skill-Supply-Chain wichtiger, nicht unwichtiger.", "tags": [ "AI Engineering", "Agent Skills", "Coding Agents", "Context Engineering", "Agent Security" ] }, { "collection": "blog", "slug": "coding-agent-save-file", "title": "Dein Coding Agent braucht eine Verfassung und ein Logbuch", "language": "de", "canonicalUrl": "https://huecki.com/blog/coding-agent-save-file/", "markdownUrl": "https://huecki.com/blog/coding-agent-save-file.md", "summary": "Coding Agents werden nicht nur durch bessere Modelle nützlich. Teams brauchen eine kleine Verfassung für Agent-Verhalten und ein kuratiertes Logbuch für Projektwissen, sonst wird Memory zu Kontext-Müll.", "tags": [ "AI Engineering", "Coding Agents", "Context Engineering", "Developer Workflow", "Agent Memory" ] }, { "collection": "blog", "slug": "debug-ai-reward-functions-like-incidents", "title": "Debug AI Reward Functions Like Production Incidents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/debug-ai-reward-functions-like-incidents/", "markdownUrl": "https://huecki.com/en/blog/debug-ai-reward-functions-like-incidents.md", "summary": "Bad reward functions should not be treated like prompt drafts. Treat them like production incidents: preserve traces, classify the failure, patch only the implicated logic, and rerun against the same controls.", "tags": [ "AI Engineering", "AI Agents", "Evals", "Developer Workflow", "Reinforcement Learning", "Agent Harness" ] }, { "collection": "blog", "slug": "dein-agent-scheiterte-vor-dem-fehler", "title": "Dein Agent scheiterte drei Schritte vor dem Fehler", "language": "de", "canonicalUrl": "https://huecki.com/blog/dein-agent-scheiterte-vor-dem-fehler/", "markdownUrl": "https://huecki.com/blog/dein-agent-scheiterte-vor-dem-fehler.md", "summary": "Bei langen Agentenläufen ist der letzte Fehler oft nur das Symptom. Der bessere Debugging-Loop sucht den frühesten kausal verantwortlichen Schritt, formuliert eine minimale Korrektur und prüft sie in einem kontrollierten Rerun.", "tags": [ "KI-Agenten", "Agent Debugging", "Observability", "Evaluation", "Developer Workflow" ] }, { "collection": "blog", "slug": "ki-code-nicht-am-diff-bewerten", "title": "Bewerte KI-Code nicht am Diff", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-code-nicht-am-diff-bewerten/", "markdownUrl": "https://huecki.com/blog/ki-code-nicht-am-diff-bewerten.md", "summary": "Besseres KI-Coding entsteht nicht primär durch bessere Prompts, sondern durch den Harness um das Modell: explizite Contracts, getrennte Builder- und Reviewer-Rollen, Belege und eine Schleife, die Fehler in bessere Spezifikationen zurückführt.", "tags": [ "AI Engineering", "KI-Agenten", "Coding Agents", "Developer Workflow", "Agent Harness", "Evals" ] }, { "collection": "blog", "slug": "ki-first-softwarearchitektur", "title": "AI-first Architektur: schneller entscheiden, sauber bleiben", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-first-softwarearchitektur/", "markdownUrl": "https://huecki.com/blog/ki-first-softwarearchitektur.md", "summary": "AI-first Architektur heißt nicht: Modell entscheidet. Es heißt: KI erzeugt Optionen, findet Risiken, verdichtet Kontext — das Team entscheidet und dokumentiert nachvollziehbar.", "tags": [ "KI", "Softwarearchitektur", "GEO", "Engineering" ] }, { "collection": "blog", "slug": "ki-generierte-ui-braucht-playtester", "title": "Deine KI-generierte UI braucht einen Playtester, keinen Screenshot-Review", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-generierte-ui-braucht-playtester/", "markdownUrl": "https://huecki.com/blog/ki-generierte-ui-braucht-playtester.md", "summary": "KI-generierte Interfaces sehen oft fertig aus, bevor sie sich korrekt verhalten. Eine GUI-Playtester-Loop schickt einen separaten Browser-Agenten in die App, protokolliert Interaktionen, speichert Screenshots und Logs, macht aus kaputten Flows reproduzierbare Bug Reports und rerunnt denselben Test nach dem Fix.", "tags": [ "AI Engineering", "KI-Agenten", "Webwright", "Playwright", "UI Testing", "Coding Agents", "Developer Workflow", "Evals" ] }, { "collection": "blog", "slug": "llm-native-developer-operational-maturity", "title": "The LLM-native developer needs more than prompts", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/llm-native-developer-operational-maturity/", "markdownUrl": "https://huecki.com/en/blog/llm-native-developer-operational-maturity.md", "summary": "The next developer skill is not writing clever prompts. It is building the operating system around LLMs: data quality, model versioning, evals, guardrails, incident response, review UX, and repo instructions agents can actually follow.", "tags": [ "AI Engineering", "LLM", "Software Architecture", "Agents", "Developer Workflow" ] }, { "collection": "blog", "slug": "llm-native-entwickler-operational-maturity", "title": "LLM-native Entwickler brauchen mehr als gute Prompts", "language": "de", "canonicalUrl": "https://huecki.com/blog/llm-native-entwickler-operational-maturity/", "markdownUrl": "https://huecki.com/blog/llm-native-entwickler-operational-maturity.md", "summary": "Die nächste Entwicklerfähigkeit ist nicht der cleverste Prompt. Es ist das Betriebssystem um LLMs herum: Datenqualität, Model-Versioning, Evals, Guardrails, Incident Response, Review-UX und Repo-Anweisungen, denen Agents wirklich folgen können.", "tags": [ "AI Engineering", "LLM", "Softwarearchitektur", "Agents", "Developer Workflow" ] }, { "collection": "blog", "slug": "natural-language-agent-harnesses", "title": "Agent harnesses should be specs, not hidden glue code", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/natural-language-agent-harnesses/", "markdownUrl": "https://huecki.com/en/blog/natural-language-agent-harnesses.md", "summary": "Natural-Language Agent Harnesses give a useful name to an important shift: the agent policy should be an inspectable document that a runtime executes, not invisible glue hidden inside controller code.", "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Evals", "Developer Workflow" ] }, { "collection": "blog", "slug": "perfect-automated-ai-agent-eval-stack", "title": "The Perfect Automated AI Eval Stack Does Not Exist", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/perfect-automated-ai-agent-eval-stack/", "markdownUrl": "https://huecki.com/en/blog/perfect-automated-ai-agent-eval-stack.md", "summary": "The reliable eval system is not one automated judge. It is a closed loop that combines portable traces, deterministic invariants, narrow semantic judges, versioned production failures, adversarial tests, and human calibration.", "tags": [ "AI Agents", "Evals", "LLM Observability", "Developer Workflow", "AI Engineering" ] }, { "collection": "blog", "slug": "prompting-2026-context-engineering", "title": "Prompting ist tot. Context zählt.", "language": "de", "canonicalUrl": "https://huecki.com/blog/prompting-2026-context-engineering/", "markdownUrl": "https://huecki.com/blog/prompting-2026-context-engineering.md", "summary": "2026 geht es nicht mehr um den einen magischen Prompt. Der bessere Ansatz: Kontext wählen, Tools und Schemas definieren, Agent-Regeln setzen und mit Evals prüfen.", "tags": [ "Prompt Engineering", "Context Engineering", "AI Agents", "KI-Workflows" ] }, { "collection": "blog", "slug": "prompting-2026-context-engineering-en", "title": "Prompting Is Dead. Context Wins.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/prompting-2026-context-engineering-en/", "markdownUrl": "https://huecki.com/en/blog/prompting-2026-context-engineering-en.md", "summary": "In 2026, good prompting is not about one magic sentence. The better approach is to curate context, define tools and schemas, set agent rules, and verify behavior with evals.", "tags": [ "Prompt Engineering", "Context Engineering", "AI Agents", "AI Workflows" ] }, { "collection": "blog", "slug": "prompting-2026-decomposition-skills-evals", "title": "Prompt Decomposition: So zerlegst du KI-Aufgaben richtig", "language": "de", "canonicalUrl": "https://huecki.com/blog/prompting-2026-decomposition-skills-evals/", "markdownUrl": "https://huecki.com/blog/prompting-2026-decomposition-skills-evals.md", "summary": "Nach Context Engineering kommt Decomposition: Entwickler sollten nicht alles in einen Prompt stopfen, sondern Aufgaben in direkte Prompts, Subtasks, Pipelines, Agent-Loops oder Skills zerlegen.", "tags": [ "Prompt Engineering", "Decomposition", "AI Agents", "Skills", "Developer Workflow" ] }, { "collection": "blog", "slug": "prompting-2026-decomposition-skills-evals-en", "title": "Prompt Decomposition: How to Break Down AI Tasks Properly", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/prompting-2026-decomposition-skills-evals-en/", "markdownUrl": "https://huecki.com/en/blog/prompting-2026-decomposition-skills-evals-en.md", "summary": "After context engineering comes decomposition: developers should stop putting everything into one prompt and instead split tasks into direct prompts, subtasks, pipelines, agent loops, or skills.", "tags": [ "Prompt Engineering", "Decomposition", "AI Agents", "Skills", "Developer Workflow" ] }, { "collection": "blog", "slug": "rag-2026-contextual-hybrid-rag", "title": "RAG 2026: Vergiss GraphRAG als Default", "language": "de", "canonicalUrl": "https://huecki.com/blog/rag-2026-contextual-hybrid-rag/", "markdownUrl": "https://huecki.com/blog/rag-2026-contextual-hybrid-rag.md", "summary": "RAG wird 2026 nicht dadurch besser, dass man GraphRAG auf alles wirft. Der robuste Default ist Contextual Hybrid RAG: saubere Ingestion, BM25 plus Embeddings, Reranking, Quellenpflicht und harte Evals.", "tags": [ "RAG", "Context Engineering", "AI Engineering", "LLM", "Evals" ] }, { "collection": "blog", "slug": "simple-graph-rag-hallucinations", "title": "Simple Graph-RAG halbiert Halluzinationen – ohne komplexes Knowledge Graph Schema", "language": "de", "canonicalUrl": "https://huecki.com/blog/simple-graph-rag-hallucinations/", "markdownUrl": "https://huecki.com/blog/simple-graph-rag-hallucinations.md", "summary": "Vector-RAG + einfacher Document-Graph halbiert Halluzinationen auf komplexen QA-Aufgaben im MoNaCo-Benchmark. Die Erkenntnis: Man braucht kein komplexes Knowledge Graph Schema – nur strukturierte Navigation über Dokumente.", "tags": [ "RAG", "GraphRAG", "Hallucinations", "Knowledge Graph", "AI Engineering", "QA" ] }, { "collection": "blog", "slug": "spec-driven-context-resets-for-coding-agents", "title": "Spec-Driven Context Resets for Coding Agents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/spec-driven-context-resets-for-coding-agents/", "markdownUrl": "https://huecki.com/en/blog/spec-driven-context-resets-for-coding-agents.md", "summary": "Long agent chats rot. A better pattern is to move decisions into small spec files, clear context between layers, and let each coding-agent session read only the artifact it needs.", "tags": [ "AI Engineering", "Coding Agents", "Context Engineering", "Spec-Driven Development", "Developer Workflow" ] }, { "collection": "blog", "slug": "spec-driven-context-resets-fuer-coding-agenten", "title": "Spec-Driven Context Resets für Coding-Agenten", "language": "de", "canonicalUrl": "https://huecki.com/blog/spec-driven-context-resets-fuer-coding-agenten/", "markdownUrl": "https://huecki.com/blog/spec-driven-context-resets-fuer-coding-agenten.md", "summary": "Lange Agenten-Chats verrotten. Besser ist es, Entscheidungen in kleine Spec-Dateien zu verschieben, zwischen den Ebenen bewusst den Kontext zu resetten und jede Coding-Agent-Session nur das lesen zu lassen, was sie wirklich braucht.", "tags": [ "AI Engineering", "Coding Agents", "Context Engineering", "Spec-Driven Development", "Developer Workflow" ] }, { "collection": "blog", "slug": "stop-judging-ai-code-by-the-diff", "title": "Stop Judging AI Code by the Diff", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-judging-ai-code-by-the-diff/", "markdownUrl": "https://huecki.com/en/blog/stop-judging-ai-code-by-the-diff.md", "summary": "Better AI coding is not mainly about better prompts. It is about the harness around the model: explicit contracts, separate builder and reviewer roles, evidence requirements, and a loop that turns failures into better specifications.", "tags": [ "AI Engineering", "AI Agents", "Coding Agents", "Developer Workflow", "Agent Harness", "Evals" ] }, { "collection": "blog", "slug": "stop-prompting-coding-agent-loop-spec", "title": "Stop Prompting Your Coding Agent. Give It a Loop.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-prompting-coding-agent-loop-spec/", "markdownUrl": "https://huecki.com/en/blog/stop-prompting-coding-agent-loop-spec.md", "summary": "The useful upgrade from prompt engineering is not a longer instruction block. It is a reusable loop spec: trigger, goal, allowed tools, verifier, terminal states, and memory rules. That is how repeated coding-agent work becomes operational instead of conversational.", "tags": [ "AI Agents", "Coding Agents", "Agent Skills", "Agent Memory", "Developer Workflow", "Verification" ] }, { "collection": "blog", "slug": "von-prompt-klappern-zu-prompt-debugging", "title": "Von 'Prompt-Klappern' zu Prompt-Debugging", "language": "de", "canonicalUrl": "https://huecki.com/blog/von-prompt-klappern-zu-prompt-debugging/", "markdownUrl": "https://huecki.com/blog/von-prompt-klappern-zu-prompt-debugging.md", "summary": "Prompt-Optimierung ist kein Suchproblem mehr — es ist ein Debugging-Problem. Contrastive Reflection liefert das Framework, um Fehler gezielt zu fangen, statt blind zu raten.", "tags": [ "AI Engineering", "Prompt Engineering", "LLM Agents", "RAG", "Evals", "KDD 2026" ] }, { "collection": "blog", "slug": "your-agent-harness-needs-a-release-process", "title": "Your Agent Harness Needs a Release Process", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/your-agent-harness-needs-a-release-process/", "markdownUrl": "https://huecki.com/en/blog/your-agent-harness-needs-a-release-process.md", "summary": "A practical field note on operating agent-harness changes like product releases: start from a trace-backed failure, change one bounded component, evaluate repeated trials and private holdouts, then promote through review with a rollback path.", "tags": [ "AI Agents", "Agent Harnesses", "Agent Evals", "Observability", "AI Governance" ] }, { "collection": "blog", "slug": "your-onboarding-is-why-your-team-is-vibe-coding", "title": "Your Onboarding Is Why Your Team Is Vibe Coding", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/your-onboarding-is-why-your-team-is-vibe-coding/", "markdownUrl": "https://huecki.com/en/blog/your-onboarding-is-why-your-team-is-vibe-coding.md", "summary": "Teams do not usually start vibe coding because developers became careless. They start because onboarding is broken: docs are stale, harnesses are undocumented, system knowledge lives in people’s heads, and AI turns missing context into plausible code and Markdown.", "tags": [ "AI Engineering", "Developer Onboarding", "Vibe Coding", "Developer Experience", "Internal Tools", "Agent Harness" ] }, { "collection": "tools", "slug": "architecture-diagram-mcp", "title": "Architecture Diagram MCP", "language": "de", "canonicalUrl": "https://huecki.com/tools/architecture-diagram-mcp/", "markdownUrl": "https://huecki.com/tools/architecture-diagram-mcp.md", "summary": "Architecture Diagram MCP erzeugt browserfähige Architekturartefakte aus Agent-Briefs. Es ist für Plattformkarten, Stakeholder-Erklärungen und reproduzierbare Diagramm-Outputs gebaut.", "tags": [ "MCP", "Architecture", "Diagrams", "Agentic Platform" ] }, { "collection": "tools", "slug": "architecture-diagram-mcp-en", "title": "Architecture Diagram MCP", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/architecture-diagram-mcp-en/", "markdownUrl": "https://huecki.com/en/tools/architecture-diagram-mcp-en.md", "summary": "Architecture Diagram MCP creates browser-ready architecture artifacts from agent briefs. It is built for platform maps, stakeholder explainers, and reproducible diagram outputs.", "tags": [ "MCP", "Architecture", "Diagrams", "Agentic Platform" ] } ] }, { "id": "buildprints", "title": "Buildprints", "titleDe": "Buildprints", "description": "Agent-readable implementation contracts with phases, prompts, evidence schemas, and validation checks.", "descriptionDe": "Agentenlesbare Implementierungsvertraege mit Phasen, Prompts, Evidenzschemas und Validierungschecks.", "urls": { "en": "https://huecki.com/en/topics/buildprints/", "de": "https://huecki.com/topics/buildprints/" }, "entries": [] } ], "posts": [ { "slug": "agent-communication-protocols-layered-stack", "title": "Agent Protocols Are Becoming a Stack, Not a Winner-Takes-All Standard", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-communication-protocols-layered-stack/", "markdownUrl": "https://huecki.com/en/blog/agent-communication-protocols-layered-stack.md", "publishedAt": "2026-06-22", "updatedAt": "2026-06-22", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "llm-native-engineering" ], "tags": [ "AI Agents", "Agent Protocols", "MCP", "A2A", "Agent Architecture" ], "summary": "The useful question is not whether MCP, A2A, ACP, agents.json, Agora, ANP, LMOS, or AGNTCY wins. The useful question is which communication boundary you are designing: discovery, tool execution, task delegation, identity, transport, or runtime negotiation.", "related": [], "sourceRefs": [] }, { "slug": "agent-eval-too-short-trajectory", "title": "Your Agent Eval Is Too Short", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-eval-too-short-trajectory/", "markdownUrl": "https://huecki.com/en/blog/agent-eval-too-short-trajectory.md", "publishedAt": "2026-07-13", "updatedAt": "2026-07-13", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Evals", "Developer Workflow", "Agent Harness", "AI Engineering" ], "summary": "A final pass/fail score hides the part of agent work that matters most: where the run started drifting, whether it noticed, and whether it recovered. The practical replacement is a trajectory eval with checkpoints, failure labels, and recovery metrics.", "related": [], "sourceRefs": [ { "title": "Long-Horizon-Terminal-Bench", "url": "https://arxiv.org/abs/2607.08964", "type": "paper" }, { "title": "Failure as a Process", "url": "https://arxiv.org/abs/2607.09510", "type": "paper" } ] }, { "slug": "agent-harness-is-a-binary", "title": "Your Agent's Harness Is a Binary Now", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-harness-is-a-binary/", "markdownUrl": "https://huecki.com/en/blog/agent-harness-is-a-binary.md", "publishedAt": "2026-06-10", "updatedAt": "2026-06-10", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Agents", "Agent Harness", "Evals", "Developer Workflow", "AI Engineering" ], "summary": "Two 2026 papers from the same research lineage quietly retire prompt engineering as a discipline. The agent's system prompt is now a binary you can version, diff, and evolve with a 200-line loop. The four metrics that actually matter are not the ones your dashboard shows.", "related": [], "sourceRefs": [] }, { "slug": "agent-harness-needs-a-behavior-map", "title": "Your Agent Harness Needs a Behavior Map", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-harness-needs-a-behavior-map/", "markdownUrl": "https://huecki.com/en/blog/agent-harness-needs-a-behavior-map.md", "publishedAt": "2026-07-17", "updatedAt": "2026-07-17", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Agent Harness", "Coding Agents", "Developer Workflow", "AI Engineering" ], "summary": "Harness Handbook points at a practical bottleneck in agent engineering: the behavior you want to change is scattered across prompts, state managers, tool calls, policy code, and tests. Build a behavior map before editing the harness.", "related": [ { "slug": "natural-language-agent-harnesses", "relation": "background on making harness policy inspectable instead of hidden in controller code" }, { "slug": "agents-dont-need-longer-prompts-they-need-harnesses", "relation": "related argument that state, evidence, and verification matter more than bigger prompts" }, { "slug": "stop-prompting-coding-agent-loop-spec", "relation": "a loop-spec workflow for coding agents once the relevant behavior is localized" } ], "sourceRefs": [ { "title": "Harness Handbook: Making Evolving Agent Harnesses Readable, Navigable, and Editable", "url": "https://arxiv.org/abs/2607.13285", "type": "paper" }, { "title": "Harness Handbook repository", "url": "https://github.com/Ruhan-Wang/Harness_Handbook", "type": "repository" }, { "title": "Harness Handbook project page", "url": "https://ruhan-wang.github.io/Harness-Handbook/", "type": "project" } ] }, { "slug": "agent-is-not-the-product-fitness-function-is", "title": "Better AI Products Need Systems, Not One Agent", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-is-not-the-product-fitness-function-is/", "markdownUrl": "https://huecki.com/en/blog/agent-is-not-the-product-fitness-function-is.md", "publishedAt": "2026-06-26", "updatedAt": "2026-07-16", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-evals", "agent-harnesses", "llm-native-engineering", "context-engineering" ], "tags": [ "AI Agents", "Agent Evals", "Self-Improvement", "Coding Agents", "AI Engineering" ], "summary": "Better AI products come from improvement systems around the agent. This guide shows how to build one with deterministic checks, narrow scoring rubrics, private holdouts, calibrated judges, and promotion gates.", "related": [ { "slug": "debug-ai-reward-functions-like-incidents", "relation": "Reward functions and fitness functions fail in similar ways." }, { "slug": "agents-dont-need-longer-prompts-they-need-harnesses", "relation": "The harness is where agent behavior becomes measurable." }, { "slug": "hermes-self-improving-agent", "relation": "A smaller practical self-review loop for repeatable skills." } ], "sourceRefs": [ { "title": "AlphaEvolve: A coding agent for scientific and algorithmic discovery", "url": "https://arxiv.org/abs/2506.13131", "type": "paper" }, { "title": "AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms", "url": "https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/", "type": "official" }, { "title": "A Self-Improving Coding Agent", "url": "https://arxiv.org/abs/2504.15228", "type": "paper" }, { "title": "CodeEvolve: an open source evolutionary coding agent for algorithmic discovery and optimization", "url": "https://arxiv.org/abs/2510.14150", "type": "paper" }, { "title": "CodeEvolve GitHub repository", "url": "https://github.com/inter-co/science-codeevolve", "type": "repository" }, { "title": "Mathematical discoveries from program search with large language models", "url": "https://www.nature.com/articles/s41586-023-06924-6", "type": "paper" } ] }, { "slug": "agent-memory-tests-measure-wrong-thing", "title": "Your Agent Memory Test Is Probably Measuring the Wrong Thing", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-memory-tests-measure-wrong-thing/", "markdownUrl": "https://huecki.com/en/blog/agent-memory-tests-measure-wrong-thing.md", "publishedAt": "2026-06-17", "updatedAt": "2026-06-17", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Memory", "Evals", "RAG", "Developer Workflow" ], "summary": "Most memory evals ask whether the agent got the final answer right. MemTrace suggests a sharper unit: one durable user fact tested across age, current state, earlier state, trajectory, and contradictory evidence. That turns memory from a vague feature into a small regression suite.", "related": [], "sourceRefs": [] }, { "slug": "agent-operating-contract-not-bigger-prompt", "title": "Your Agent Needs an Operating Contract, Not a Bigger Prompt", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-operating-contract-not-bigger-prompt/", "markdownUrl": "https://huecki.com/en/blog/agent-operating-contract-not-bigger-prompt.md", "publishedAt": "2026-07-06", "updatedAt": "2026-07-06", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Agents", "Coding Agents", "Agent Workflow", "Verification", "Agent Skills", "Developer Workflow" ], "summary": "The serious agent pattern is no longer bigger prompts and more encouragement. It is an operating contract: measurable goal, bounded tools, context sources, verifier evidence, review notes, rollback path, and a skill update when the run teaches you something.", "related": [ { "slug": "stop-prompting-coding-agent-loop-spec", "relation": "Loop specs define the recurring workflow; operating contracts attach evidence, review, rollback, and learning to each run." }, { "slug": "ai-agent-permission-layer-not-popups", "relation": "Permissions are one part of the operating contract around an agent's tool use." }, { "slug": "failed-agent-runs-should-become-skills", "relation": "A failed run should update the skill only when the evidence explains what changed." } ], "sourceRefs": [ { "title": "TLDR AI, July 3 2026", "url": "https://tldr.tech/ai/2026-07-03", "type": "curator" }, { "title": "AIEWF Daily Dispatch: Locomotives", "url": "https://www.latent.space/p/aiewf-daily-dispatch-locomotives", "type": "newsletter" }, { "title": "SkillOpt repository", "url": "https://github.com/microsoft/SkillOpt", "type": "repository" }, { "title": "Using Local Coding Agents", "url": "https://magazine.sebastianraschka.com/p/using-local-coding-agents", "type": "article" }, { "title": "Huecki AI radar, 2026-07-06", "url": "https://huecki.com", "type": "internal-radar" } ] }, { "slug": "agent-plugins-npm-fuer-agent-verhalten", "title": "Agent Plugins sind npm für Agent-Verhalten — aber ohne Lockfile", "language": "de", "canonicalUrl": "https://huecki.com/blog/agent-plugins-npm-fuer-agent-verhalten/", "markdownUrl": "https://huecki.com/blog/agent-plugins-npm-fuer-agent-verhalten.md", "publishedAt": "2026-08-06", "updatedAt": "2026-08-06", "topic": "AI Agent Infrastructure", "agentTopicIds": [ "agent-security", "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Agent Plugins", "Agent Skills", "MCP", "Codex", "Cursor", "VS Code", "AI Security" ], "summary": "Agent Plugins machen aus Skills und MCP-Konfigurationen ein portables Paket. Der Guide zeigt den kleinsten Aufbau, die Integration in mehrere Clients sowie die fehlenden Produktionskontrollen für Permissions, Updates, Evals und Rollback.", "related": [ { "slug": "agent-skills-need-a-supply-chain-gate", "relation": "Zeigt, warum installierte Agent-Fähigkeiten vor der Aktivierung geprüft werden müssen." }, { "slug": "automatisch-generierte-agent-skills-supply-chain", "relation": "Vertieft Provenance, Scanning und Freigabe automatisch erzeugter Skills." }, { "slug": "mcp-tools-permission-labels", "relation": "Erklärt, warum MCP-Tools verständliche Berechtigungslabels brauchen." } ], "sourceRefs": [ { "title": "Agent Plugins overview", "url": "https://agent-plugins.org/", "type": "official standard documentation" }, { "title": "Agent Plugins Specification 1.0.0", "url": "https://agent-plugins.org/specification", "type": "specification" }, { "title": "Compatible Agent Plugins clients", "url": "https://agent-plugins.org/compatible-clients", "type": "official compatibility matrix" }, { "title": "Build an Agent Plugin", "url": "https://agent-plugins.org/plugin-authors", "type": "official guide" }, { "title": "Implement an Agent Plugins client", "url": "https://agent-plugins.org/client-implementers", "type": "official guide" }, { "title": "Package plugins for ChatGPT and Codex", "url": "https://developers.openai.com/plugins/build/plugins", "type": "vendor documentation" }, { "title": "Agent plugins in VS Code", "url": "https://code.visualstudio.com/docs/agent-customization/agent-plugins", "type": "vendor documentation" }, { "title": "About GitHub Copilot plugins", "url": "https://docs.github.com/en/copilot/concepts/agents/about-plugins", "type": "vendor documentation" }, { "title": "Cursor plugin marketplace", "url": "https://cursor.com/blog/marketplace", "type": "vendor announcement" }, { "title": "AWS Agent Plugins", "url": "https://github.com/awslabs/agent-plugins", "type": "official repository" }, { "title": "Agent Skills overview", "url": "https://agentskills.io/home", "type": "open standard documentation" }, { "title": "Model Context Protocol specification", "url": "https://modelcontextprotocol.io/specification/2025-11-25", "type": "specification" } ] }, { "slug": "agent-skills-qualitaets-gate", "title": "Agent Skills sind kein Markdown. Sie brauchen ein Qualitäts-Gate.", "language": "de", "canonicalUrl": "https://huecki.com/blog/agent-skills-qualitaets-gate/", "markdownUrl": "https://huecki.com/blog/agent-skills-qualitaets-gate.md", "publishedAt": "2026-08-11", "updatedAt": "2026-08-11", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-security", "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Agent Skills", "Agent Evals", "AI Security", "Developer Workflow" ], "summary": "Ein praktisches Qualitäts-Gate für Agent Skills: Nutzen gegen eine No-Skill-Baseline messen, Aktivierung und Trajektorie prüfen, Rechte außerhalb des Skill-Texts binden und nur versionierte, überprüfbare Kandidaten promoten.", "related": [ { "slug": "automatisch-generierte-agent-skills-supply-chain", "relation": "Zeigt, warum automatisch erzeugte Skills Provenance, Quarantäne und eine Registry brauchen." }, { "slug": "agent-plugins-npm-fuer-agent-verhalten", "relation": "Ordnet Skills und Plugins als versionierte Verhaltensabhängigkeiten ein." }, { "slug": "your-agent-harness-needs-a-release-process", "relation": "Überträgt Release Engineering auf Änderungen am gesamten Agent Harness." } ], "sourceRefs": [ { "title": "SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks", "url": "https://arxiv.org/abs/2602.12670", "type": "paper" }, { "title": "SWE-Skills-Bench: Do Agent Skills Actually Help in Real-World Software Engineering?", "url": "https://arxiv.org/abs/2603.15401", "type": "paper" }, { "title": "SkillLearnBench: Benchmarking Continual Learning Methods for Agent Skill Generation", "url": "https://arxiv.org/abs/2604.20087", "type": "paper" }, { "title": "Under the Hood of SKILL.md: Semantic Supply-chain Attacks", "url": "https://arxiv.org/abs/2605.11418", "type": "paper" }, { "title": "SkillGate: Cost Efficient Runtime Malicious Skill File Detection", "url": "https://arxiv.org/abs/2607.25619", "type": "paper" }, { "title": "ClawHub Security Signals", "url": "https://arxiv.org/abs/2606.01494", "type": "paper" }, { "title": "Agent Skills specification", "url": "https://github.com/agentskills/agentskills/blob/main/docs/specification.mdx", "type": "specification" }, { "title": "Cisco AI Skill Scanner", "url": "https://github.com/cisco-ai-defense/skill-scanner", "type": "repository" } ] }, { "slug": "agent-state-machines-seatbelts", "title": "Give Your Agent Seatbelts, Not a Longer Prompt", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agent-state-machines-seatbelts/", "markdownUrl": "https://huecki.com/en/blog/agent-state-machines-seatbelts.md", "publishedAt": "2026-05-25", "updatedAt": "2026-05-25", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Coding Agents", "Agent Harness", "Developer Workflow", "AI Safety" ], "summary": "When an agent keeps jumping from planning to editing to testing at the wrong time, the fix is not usually another paragraph of system prompt. Put the workflow into explicit states, give each state a tiny tool policy, and make phase changes visible.", "related": [], "sourceRefs": [] }, { "slug": "agenten-brauchen-runtime-contracts", "title": "Agenten brauchen Runtime Contracts, nicht längere Prompts", "language": "de", "canonicalUrl": "https://huecki.com/blog/agenten-brauchen-runtime-contracts/", "markdownUrl": "https://huecki.com/blog/agenten-brauchen-runtime-contracts.md", "publishedAt": "2026-06-07", "updatedAt": "2026-06-07", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Developer Workflow", "KI-Workflows", "Evals" ], "summary": "Bessere Prompts machen Agenten nicht automatisch zuverlässig. Entwickler brauchen Runtime Contracts: explizite Verträge dafür, welche Tools ein Agent nutzen darf, was er erinnern darf, wann er stoppen muss und wie seine Behauptungen geprüft werden.", "related": [], "sourceRefs": [] }, { "slug": "agents-dont-need-longer-prompts-they-need-harnesses", "title": "Agents Don’t Need Longer Prompts. They Need Harnesses.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses/", "markdownUrl": "https://huecki.com/en/blog/agents-dont-need-longer-prompts-they-need-harnesses.md", "publishedAt": "2026-05-20", "updatedAt": "2026-05-20", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "agent-security", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Coding Agents", "Evals", "Developer Workflow" ], "summary": "The arXiv survey Code as Agent Harness names the next shift in agent engineering: code is not only what agents generate. It is becoming the executable, inspectable, stateful runtime that makes agents reliable.", "related": [], "sourceRefs": [] }, { "slug": "agents-md-coding-agent-harness", "title": "AGENTS.md reicht nicht: Dein Coding Agent braucht einen Harness", "language": "de", "canonicalUrl": "https://huecki.com/blog/agents-md-coding-agent-harness/", "markdownUrl": "https://huecki.com/blog/agents-md-coding-agent-harness.md", "publishedAt": "2026-05-19", "updatedAt": "2026-05-19", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "agent-security", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ], "summary": "Ein Coding Agent wird nicht durch einen magischen Prompt zuverlässig. Er braucht einen Harness: AGENTS.md, Skills, Tool-Permissions, Hooks und Evals, die merken, wenn sich sein Verhalten verschiebt.", "related": [], "sourceRefs": [] }, { "slug": "agents-md-coding-agent-harness-en", "title": "AGENTS.md is not enough: your coding agent needs a harness", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en/", "markdownUrl": "https://huecki.com/en/blog/agents-md-coding-agent-harness-en.md", "publishedAt": "2026-05-19", "updatedAt": "2026-05-19", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "agent-security", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Evals", "Developer Workflow" ], "summary": "A coding agent is not made reliable by one magic prompt. It needs a harness: AGENTS.md, skills, tool permissions, hooks, and evals that catch behavior drift.", "related": [], "sourceRefs": [] }, { "slug": "agents-md-control-surface", "title": "AGENTS.md ist kein Kontext. Es ist eine Steuerfläche.", "language": "de", "canonicalUrl": "https://huecki.com/blog/agents-md-control-surface/", "markdownUrl": "https://huecki.com/blog/agents-md-control-surface.md", "publishedAt": "2026-06-08", "updatedAt": "2026-06-08", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Context Engineering", "Developer Workflow" ], "summary": "Die überraschende Lektion aus AGENTS.md-Benchmarks ist nicht, dass Kontextdateien nutzlos sind. Sie verändern Agent-Verhalten, manchmal in Richtung teurerer und weniger nützlicher Arbeit. Behandle sie als Steuerfläche, nicht als Repo-Handbuch.", "related": [], "sourceRefs": [] }, { "slug": "agents-md-control-surface-en", "title": "AGENTS.md Is Not Context. It Is a Control Surface.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-md-control-surface-en/", "markdownUrl": "https://huecki.com/en/blog/agents-md-control-surface-en.md", "publishedAt": "2026-06-08", "updatedAt": "2026-06-08", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "Coding Agents", "AGENTS.md", "Context Engineering", "Developer Workflow" ], "summary": "The surprising lesson from AGENTS.md benchmarks is not that context files are useless. It is that they change agent behavior, sometimes into more expensive and less useful work. Treat them as a control surface, not a repo manual.", "related": [], "sourceRefs": [] }, { "slug": "agents-need-exit-conditions", "title": "Agents Don’t Need ‘Keep Going’. They Need Exit Conditions.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/agents-need-exit-conditions/", "markdownUrl": "https://huecki.com/en/blog/agents-need-exit-conditions.md", "publishedAt": "2026-05-26", "updatedAt": "2026-05-26", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Agent Harness", "Developer Workflow", "AI Engineering", "Evals" ], "summary": "The useful lesson behind Claude Code /goal is not that agents can run forever. It is that long-running agent work needs an explicit, observable exit condition: what proves done, what stays in scope, and when to stop blocked.", "related": [], "sourceRefs": [] }, { "slug": "ai-agents-need-evidence-before-clicking", "title": "AI Agents Need Evidence Before They Click", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-agents-need-evidence-before-clicking/", "markdownUrl": "https://huecki.com/en/blog/ai-agents-need-evidence-before-clicking.md", "publishedAt": "2026-05-21", "updatedAt": "2026-05-21", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-harnesses", "agent-security", "context-engineering" ], "tags": [ "AI Agents", "Multimodal AI", "Browser Agents", "AI Safety", "Developer Workflow" ], "summary": "When an agent clicks, sends, pays, deletes, or extracts data, the critical truth cannot live only in model prose. Put a small evidence gate before risky tool calls: predicate, evidence type, source, decision.", "related": [], "sourceRefs": [] }, { "slug": "ai-built-ui-needs-a-playtester", "title": "Your AI-Built UI Needs a Playtester, Not a Screenshot Review", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-built-ui-needs-a-playtester/", "markdownUrl": "https://huecki.com/en/blog/ai-built-ui-needs-a-playtester.md", "publishedAt": "2026-05-28", "updatedAt": "2026-05-28", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "AI Agents", "Webwright", "Playwright", "UI Testing", "Coding Agents", "Developer Workflow", "Evals" ], "summary": "AI-generated interfaces often look finished before they behave correctly. A GUI playtester loop uses a separate browser agent to interact with the artifact, record screenshots and action logs, turn broken flows into reproducible bug reports, and rerun the same script after repairs.", "related": [], "sourceRefs": [] }, { "slug": "ai-first-software-architecture", "title": "AI-first Architecture: Faster Decisions, Still in Control", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-first-software-architecture/", "markdownUrl": "https://huecki.com/en/blog/ai-first-software-architecture.md", "publishedAt": "2026-04-29", "updatedAt": "2026-05-12", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-security", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI", "Software Architecture", "GEO", "Engineering" ], "summary": "AI-first architecture does not mean the model decides. It means AI generates options, finds risks, compresses context, and the team makes a traceable decision.", "related": [], "sourceRefs": [] }, { "slug": "ai-self-check-always-finds-something", "title": "Stop Asking AI to Critically Self-Check", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-self-check-always-finds-something/", "markdownUrl": "https://huecki.com/en/blog/ai-self-check-always-finds-something.md", "publishedAt": "2026-05-21", "updatedAt": "2026-05-21", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Prompt Engineering", "Evals", "AI Safety", "Developer Workflow" ], "summary": "Open-ended instructions like “critically self-check this” accidentally reward the model for producing criticism. The fix is not less review. It is calibrated review: explicit criteria, PASS_NO_CHANGE, evidence per finding, severity thresholds, and a tiny change budget.", "related": [], "sourceRefs": [] }, { "slug": "ai-self-reflection-defends-first-answer", "title": "Your AI Agent Is Not Reflecting. It Is Defending Its First Answer", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-self-reflection-defends-first-answer/", "markdownUrl": "https://huecki.com/en/blog/ai-self-reflection-defends-first-answer.md", "publishedAt": "2026-07-17", "updatedAt": "2026-07-17", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-evals", "context-engineering", "llm-native-engineering" ], "tags": [ "AI Agents", "Multi-Agent Systems", "Self-Reflection", "LLM Judges", "Agent Architecture" ], "summary": "Asking one agent to reconsider its answer often produces a more confident defense of the same mistake. A bounded challenger-and-judge loop can create real alternatives, but only if disagreement, stopping, and judge bias are engineered explicitly.", "related": [ { "slug": "ai-self-check-always-finds-something", "relation": "Explains why open-ended self-review also creates unnecessary changes." }, { "slug": "agent-is-not-the-product-fitness-function-is", "relation": "Shows how external evals decide whether an agent workflow actually improved." }, { "slug": "agents-need-exit-conditions", "relation": "Covers the stopping rules required by iterative agent loops." } ], "sourceRefs": [ { "title": "Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate", "url": "https://arxiv.org/abs/2305.19118", "type": "paper" }, { "title": "Improving Factuality and Reasoning in Language Models through Multiagent Debate", "url": "https://arxiv.org/abs/2305.14325", "type": "paper" }, { "title": "Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena", "url": "https://arxiv.org/abs/2306.05685", "type": "paper" } ] }, { "slug": "ai-slop-gate-after-tests-and-lint", "title": "Put an AI Slop Gate After Tests and Lint", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/ai-slop-gate-after-tests-and-lint/", "markdownUrl": "https://huecki.com/en/blog/ai-slop-gate-after-tests-and-lint.md", "publishedAt": "2026-05-30", "updatedAt": "2026-05-30", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "Coding Agents", "Developer Workflow", "Code Quality", "Evals", "Agent Harness" ], "summary": "Tests tell you whether behavior still works. Linters tell you whether code is syntactically and stylistically acceptable. An AI-slop gate catches the residue coding agents leave behind: fake comments, swallowed errors, any-casts, duplicated helpers, TODO stubs, and dead code.", "related": [], "sourceRefs": [] }, { "slug": "automatisch-generierte-agent-skills-supply-chain", "title": "Automatisch generierte Agent Skills brauchen eine Supply Chain", "language": "de", "canonicalUrl": "https://huecki.com/blog/automatisch-generierte-agent-skills-supply-chain/", "markdownUrl": "https://huecki.com/blog/automatisch-generierte-agent-skills-supply-chain.md", "publishedAt": "2026-06-16", "updatedAt": "2026-06-16", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-security", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "Agent Skills", "Coding Agents", "Context Engineering", "Agent Security" ], "summary": "Das OpenClaw-Skill-Paper ist ein starkes Signal: Agent Skills werden nicht nur manuell geschrieben, sondern aus Trajektorien, Skill-Bäumen und Transfer-Evals gelernt. Aber genau das macht eine Skill-Supply-Chain wichtiger, nicht unwichtiger.", "related": [], "sourceRefs": [] }, { "slug": "client-projekt-agenten-standup", "title": "gib jedem Kundenprojekt einen kleinen Agenten", "language": "de", "canonicalUrl": "https://huecki.com/blog/client-projekt-agenten-standup/", "markdownUrl": "https://huecki.com/blog/client-projekt-agenten-standup.md", "publishedAt": "2026-05-19", "updatedAt": "2026-05-19", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Freelancing", "Client Work", "Automation", "Workflow Design" ], "summary": "Der nützliche Move ist nicht ein Mega-Assistent für alle Kunden. Gib jedem Kundenprojekt einen kleinen, isolierten Agenten mit eigener Erinnerung, Aufgabenliste, Preview-URL-Gewohnheit und langweiligem Daily Standup.", "related": [], "sourceRefs": [] }, { "slug": "client-site-domain-agents", "title": "give every client project a tiny agent", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/client-site-domain-agents/", "markdownUrl": "https://huecki.com/en/blog/client-site-domain-agents.md", "publishedAt": "2026-05-19", "updatedAt": "2026-05-19", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Freelancing", "Client Work", "Automation", "Workflow Design" ], "summary": "The useful move is not one mega assistant for all client work. Give each client project a small, isolated agent with its own memory, tasks, preview URL habit, and boring daily standup.", "related": [], "sourceRefs": [] }, { "slug": "coding-agent-command-composition-risk", "title": "Your Coding Agent Can Be Tricked by Boring Shell Commands", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/coding-agent-command-composition-risk/", "markdownUrl": "https://huecki.com/en/blog/coding-agent-command-composition-risk.md", "publishedAt": "2026-07-07", "updatedAt": "2026-07-07", "topic": "AI Agent Security", "agentTopicIds": [ "agent-security", "context-engineering" ], "tags": [ "AI Agents", "Coding Agents", "Security", "Developer Workflow", "Failure Mode" ], "summary": "The MOSAIC paper shifts the coding-agent security question from hostile prompts to command traces. The practical move is to audit producer-consumer state across shell commands before generated state crosses into privileged work.", "related": [], "sourceRefs": [] }, { "slug": "coding-agent-instruktionen-testen", "title": "Teste deine Coding-Agent-Instruktionen wie Code", "language": "de", "canonicalUrl": "https://huecki.com/blog/coding-agent-instruktionen-testen/", "markdownUrl": "https://huecki.com/blog/coding-agent-instruktionen-testen.md", "publishedAt": "2026-06-22", "updatedAt": "2026-06-22", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals" ], "tags": [ "Coding Agents", "AI Engineering", "Developer Workflow", "AGENTS.md", "Evals" ], "summary": "Eine gute AGENTS.md ist kein Prompt-Dokument, das man einmal schreibt. Sie ist ein kleines Betriebshandbuch fuer den Agenten. Und Betriebshandbuecher werden besser, wenn man sie gegen konkrete Fehler testet.", "related": [], "sourceRefs": [] }, { "slug": "coding-agent-save-file", "title": "Dein Coding Agent braucht eine Verfassung und ein Logbuch", "language": "de", "canonicalUrl": "https://huecki.com/blog/coding-agent-save-file/", "markdownUrl": "https://huecki.com/blog/coding-agent-save-file.md", "publishedAt": "2026-06-15", "updatedAt": "2026-06-15", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "llm-native-engineering" ], "tags": [ "AI Engineering", "Coding Agents", "Context Engineering", "Developer Workflow", "Agent Memory" ], "summary": "Coding Agents werden nicht nur durch bessere Modelle nützlich. Teams brauchen eine kleine Verfassung für Agent-Verhalten und ein kuratiertes Logbuch für Projektwissen, sonst wird Memory zu Kontext-Müll.", "related": [], "sourceRefs": [] }, { "slug": "coding-agents-need-hardened-harness-evals", "title": "Coding Agents Need Hardened Harness Evals", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/coding-agents-need-hardened-harness-evals/", "markdownUrl": "https://huecki.com/en/blog/coding-agents-need-hardened-harness-evals.md", "publishedAt": "2026-08-05", "updatedAt": "2026-08-05", "topic": "AI Agent Security", "agentTopicIds": [ "agent-harnesses", "agent-security", "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Coding Agents", "Agent Evals", "Security", "Developer Workflow" ], "summary": "Permissive coding-agent benchmarks hide a boring production truth: security policy changes agent behavior. Small teams should run the same task suite under nested hardening levels and separate model failures from tasks the policy made impossible.", "related": [ { "slug": "benchmark-scores-are-not-agent-validation", "relation": "Explains why model benchmarks are only input evidence, not deployment validation." }, { "slug": "mcp-tools-permission-labels", "relation": "Shows the same permission-label habit for tool-calling agents." }, { "slug": "coding-agent-command-composition-risk", "relation": "Covers command-trace risk once agents can run shell workflows." } ], "sourceRefs": [ { "title": "Permission Denied: Policy-Graded Evaluation of Coding Agents in Hardened Environments", "url": "https://arxiv.org/abs/2608.02670", "type": "paper" }, { "title": "CUADebug: Diagnosing and Repairing Computer-Use Agent Failures", "url": "https://arxiv.org/abs/2608.02643", "type": "paper" }, { "title": "ExplainBench: Evaluating Code Explanations from Agents", "url": "https://huggingface.co/papers/2607.26451", "type": "paper" } ] }, { "slug": "debug-ai-reward-functions-like-incidents", "title": "Debug AI Reward Functions Like Production Incidents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/debug-ai-reward-functions-like-incidents/", "markdownUrl": "https://huecki.com/en/blog/debug-ai-reward-functions-like-incidents.md", "publishedAt": "2026-05-30", "updatedAt": "2026-05-30", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "AI Agents", "Evals", "Developer Workflow", "Reinforcement Learning", "Agent Harness" ], "summary": "Bad reward functions should not be treated like prompt drafts. Treat them like production incidents: preserve traces, classify the failure, patch only the implicated logic, and rerun against the same controls.", "related": [], "sourceRefs": [] }, { "slug": "dein-agent-scheiterte-vor-dem-fehler", "title": "Dein Agent scheiterte drei Schritte vor dem Fehler", "language": "de", "canonicalUrl": "https://huecki.com/blog/dein-agent-scheiterte-vor-dem-fehler/", "markdownUrl": "https://huecki.com/blog/dein-agent-scheiterte-vor-dem-fehler.md", "publishedAt": "2026-07-22", "updatedAt": "2026-07-22", "topic": "KI-Agent Reliability", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "KI-Agenten", "Agent Debugging", "Observability", "Evaluation", "Developer Workflow" ], "summary": "Bei langen Agentenläufen ist der letzte Fehler oft nur das Symptom. Der bessere Debugging-Loop sucht den frühesten kausal verantwortlichen Schritt, formuliert eine minimale Korrektur und prüft sie in einem kontrollierten Rerun.", "related": [ { "slug": "agent-failure-vocabulary", "relation": "Eine Failure Vocabulary benennt wiederkehrende Fehler; Root-Cause Attribution lokalisiert den verantwortlichen Schritt im einzelnen Lauf." }, { "slug": "failed-agent-runs-should-become-skills", "relation": "Erst eine belegte Diagnose sollte aus einem fehlgeschlagenen Lauf eine dauerhafte Agent-Regel machen." }, { "slug": "perfect-automated-ai-agent-eval-stack", "relation": "Kontrollierte Reruns und gespeicherte Regression Cases verbinden Debugging mit Evaluation." } ], "sourceRefs": [ { "title": "AgentDebugX: An Open-Source Toolkit for Failure Observability, Attribution, and Recovery in LLM Agents", "url": "https://arxiv.org/abs/2607.18754", "type": "paper" }, { "title": "AgentDebugX GitHub repository", "url": "https://github.com/AgentDebugX/AgentDebugX", "type": "code" }, { "title": "AgentDebugX project and demo", "url": "https://www.agentdebugx.com/", "type": "project" } ] }, { "slug": "dein-ai-agent-lernt-nichts-aus-seinen-runs", "title": "Dein AI Agent lernt nichts aus seinen Runs", "language": "de", "canonicalUrl": "https://huecki.com/blog/dein-ai-agent-lernt-nichts-aus-seinen-runs/", "markdownUrl": "https://huecki.com/blog/dein-ai-agent-lernt-nichts-aus-seinen-runs.md", "publishedAt": "2026-08-06", "updatedAt": "2026-08-06", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Knowledge Flywheel", "Agent Memory", "Agent Skills", "Self-Improving Agents", "Agent Evals" ], "summary": "Ein Agent verbessert sich nicht, nur weil seine Runs gespeichert werden. Ein Knowledge Flywheel extrahiert aus mehreren Runs belegte Lessons, prüft Widersprüche, versioniert das Ergebnis und liefert dem nächsten Agenten nur das Wissen, das zu seinem Task passt.", "related": [ { "slug": "should-ai-agent-remember-what-it-learned", "relation": "Erklärt, warum persistente Agent-Learnings wie Deployments behandelt werden müssen." }, { "slug": "coding-agent-instruktionen-testen", "relation": "Zeigt, wie wiederverwendbare Agent-Instruktionen mit ausführbaren Evals geprüft werden." }, { "slug": "hermes-selbstverbessernder-agent", "relation": "Zeigt einen praktischen selbstverbessernden Agenten, dessen dauerhafte Skills einen kontrollierten Lern-Loop brauchen." } ], "sourceRefs": [ { "title": "Knowledge-Centric Self-Improvement", "url": "https://arxiv.org/abs/2607.19592", "type": "paper" }, { "title": "Knowledge-Centric Self-Improvement project and results", "url": "https://recursive-knowledge.github.io/knowledge-centric-self-improvement/", "type": "project page" }, { "title": "KSI open-source implementation", "url": "https://github.com/recursive-knowledge/KSI", "type": "source code" }, { "title": "Trace2Skill: Distill Trajectory-Local Lessons into Transferable Agent Skills", "url": "https://arxiv.org/abs/2603.25158", "type": "paper" }, { "title": "Agent KB: Leveraging Cross-Domain Experience for Agentic Problem Solving", "url": "https://arxiv.org/abs/2507.06229", "type": "paper" }, { "title": "Agent Workflow Memory", "url": "https://arxiv.org/abs/2409.07429", "type": "paper" }, { "title": "SkillWeaver: Web Agents can Self-Improve by Discovering and Honing Skills", "url": "https://arxiv.org/abs/2504.07079", "type": "paper" }, { "title": "ConsistencyGate: Preventing Memory Contamination in LLM Agents", "url": "https://arxiv.org/abs/2607.22962", "type": "paper" }, { "title": "Memory Contagion: Cross-Temporal Propagation of Evaluator Bias via Agent Memory", "url": "https://arxiv.org/abs/2606.23195", "type": "paper" }, { "title": "The llms.txt proposal", "url": "https://llmstxt.org/", "type": "proposal" }, { "title": "Model Context Protocol Resources", "url": "https://modelcontextprotocol.io/specification/2025-03-26/server/resources", "type": "specification" } ] }, { "slug": "graph-engineering-routing-wird-produktverhalten", "title": "Graph Engineering beginnt dort, wo Routing zum Produktverhalten wird", "language": "de", "canonicalUrl": "https://huecki.com/blog/graph-engineering-routing-wird-produktverhalten/", "markdownUrl": "https://huecki.com/blog/graph-engineering-routing-wird-produktverhalten.md", "publishedAt": "2026-07-26", "updatedAt": "2026-07-26", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "agent-evals" ], "tags": [ "Graph Engineering", "LangGraph", "AI Agents", "Agent Workflows", "Human-in-the-Loop", "AI Engineering" ], "summary": "Graph Engineering ist nicht das Zeichnen komplexer Agentendiagramme. Es beginnt dann, wenn Retry, Eskalation, Evidenzprüfung oder menschliche Freigabe sichtbares und testbares Produktverhalten werden.", "related": [ { "slug": "agent-state-machines-seatbelts", "relation": "Zeigt, wie explizite Zustände unerlaubte Phasenwechsel verhindern." }, { "slug": "next-prompt-is-a-workflow", "relation": "Erklärt den Wechsel vom Prompt zur beobachtbaren Orchestrierung." }, { "slug": "harness-engineering-field-guide", "relation": "Ordnet Routing, State, Verification und Recovery in den gesamten Agent Harness ein." }, { "slug": "rag-evidence-set-contracts", "relation": "Vertieft Evidenz-Gates für Retrieval-Workflows." } ], "sourceRefs": [ { "title": "Graph-Based Agentic AI with LangGraph", "url": "https://arxiv.org/abs/2607.19297", "type": "paper" }, { "title": "Ancillary source package", "url": "https://arxiv.org/e-print/2607.19297", "type": "source code" }, { "title": "LangGraph overview", "url": "https://docs.langchain.com/oss/python/langgraph/overview", "type": "official documentation" }, { "title": "LangGraph Graph API", "url": "https://docs.langchain.com/oss/python/langgraph/graph-api", "type": "official documentation" }, { "title": "LangGraph persistence", "url": "https://docs.langchain.com/oss/python/langgraph/persistence", "type": "official documentation" }, { "title": "LangGraph interrupts", "url": "https://docs.langchain.com/oss/python/langgraph/interrupts", "type": "official documentation" }, { "title": "LangGraph Functional API", "url": "https://docs.langchain.com/oss/python/langgraph/functional-api", "type": "official documentation" }, { "title": "ReAct", "url": "https://arxiv.org/abs/2210.03629", "type": "paper" }, { "title": "DSPy", "url": "https://arxiv.org/abs/2310.03714", "type": "paper" }, { "title": "How we built our multi-agent research system", "url": "https://www.anthropic.com/engineering/multi-agent-research-system", "type": "engineering report" } ] }, { "slug": "harness-engineering-field-guide", "title": "Harness Engineering Is Systems Engineering for AI Agents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/harness-engineering-field-guide/", "markdownUrl": "https://huecki.com/en/blog/harness-engineering-field-guide.md", "publishedAt": "2026-07-18", "updatedAt": "2026-07-18", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "agent-evals", "agent-security", "agent-harnesses" ], "tags": [ "Harness Engineering", "AI Agents", "Agent Evals", "AI Security", "AI Engineering" ], "summary": "Harness engineering is the work of turning a probabilistic model into a controlled system. This field guide maps the twelve engineering surfaces around the model and shows how to evaluate and evolve them.", "related": [ { "slug": "agents-dont-need-longer-prompts-they-need-harnesses", "relation": "Introduces code as the executable and verifiable substrate around agents." }, { "slug": "agent-harness-needs-a-behavior-map", "relation": "Shows how to localize a harness behavior before changing its implementation." }, { "slug": "should-ai-agent-remember-what-it-learned", "relation": "Explains why persistent harness changes require promotion gates and rollback." }, { "slug": "your-ai-agent-bill-is-written-by-the-harness", "relation": "Explores the token and latency economics controlled by orchestration." } ], "sourceRefs": [ { "title": "Agent Harness Engineering: A Survey", "url": "https://openreview.net/forum?id=eONq7FdiHa", "type": "survey" }, { "title": "Harness-Bench: Measuring Harness Effects across Models in Realistic Agent Workflows", "url": "https://arxiv.org/abs/2605.27922", "type": "paper" }, { "title": "Code as Agent Harness", "url": "https://arxiv.org/abs/2605.18747", "type": "survey" }, { "title": "Auditing Agent Harness Safety", "url": "https://arxiv.org/abs/2605.14271", "type": "paper" }, { "title": "Rethinking the Evaluation of Harness Evolution for Agents", "url": "https://arxiv.org/abs/2607.12227", "type": "paper" }, { "title": "Effective context engineering for AI agents", "url": "https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents", "type": "engineering guide" }, { "title": "Writing effective tools for AI agents", "url": "https://www.anthropic.com/engineering/writing-tools-for-agents", "type": "engineering guide" }, { "title": "Demystifying evals for AI agents", "url": "https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents", "type": "engineering guide" }, { "title": "AgentDojo", "url": "https://arxiv.org/abs/2406.13352", "type": "paper" }, { "title": "Model Context Protocol Security Best Practices", "url": "https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices", "type": "specification guidance" } ] }, { "slug": "hermes-selbstverbessernder-agent", "title": "Hermes Agent: Self-Review statt One-Shot", "language": "de", "canonicalUrl": "https://huecki.com/blog/hermes-selbstverbessernder-agent/", "markdownUrl": "https://huecki.com/blog/hermes-selbstverbessernder-agent.md", "publishedAt": "2026-05-11", "updatedAt": "2026-05-12", "topic": "KI-Agenten-Workflows", "agentTopicIds": [ "context-engineering", "agent-evals" ], "tags": [ "Hermes Agent", "KI-Agenten", "Self-Improvement", "Nous Research" ], "summary": "Hermes wird interessant, wenn ein Agent nicht nur liefert, sondern die eigene Arbeit reviewed: ausführen, messen, kritisieren, Skill umbauen, nochmal laufen lassen. Der Nutzen entsteht vor allem bei wiederholbaren Workflows.", "related": [], "sourceRefs": [] }, { "slug": "hermes-self-improving-agent", "title": "Hermes Agent: Self-Review Instead of One-Shot Output", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/hermes-self-improving-agent/", "markdownUrl": "https://huecki.com/en/blog/hermes-self-improving-agent.md", "publishedAt": "2026-05-11", "updatedAt": "2026-05-12", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "agent-evals" ], "tags": [ "Hermes Agent", "AI Agents", "Self-Improvement", "Nous Research" ], "summary": "Hermes gets interesting when an agent does not only produce output, but reviews the run: execute, measure, critique, rewrite the skill, and test again. The loop pays off mainly for repeatable workflows.", "related": [], "sourceRefs": [] }, { "slug": "ki-agenten-belege-vor-dem-klick", "title": "KI-Agenten brauchen Belege, bevor sie klicken", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-agenten-belege-vor-dem-klick/", "markdownUrl": "https://huecki.com/blog/ki-agenten-belege-vor-dem-klick.md", "publishedAt": "2026-05-21", "updatedAt": "2026-05-21", "topic": "KI-Agent Workflows", "agentTopicIds": [ "agent-security", "context-engineering" ], "tags": [ "KI-Agenten", "Multimodal AI", "Browser Agents", "AI Safety", "Developer Workflow" ], "summary": "Wenn ein Agent klickt, sendet, kauft oder Daten extrahiert, darf die entscheidende Wahrheit nicht nur aus Modell-Prosa kommen. Baue vor riskanten Tool Calls ein kleines Evidenz-Gate: Predicate, Belegtyp, Quelle, Entscheidung.", "related": [], "sourceRefs": [] }, { "slug": "ki-code-nicht-am-diff-bewerten", "title": "Bewerte KI-Code nicht am Diff", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-code-nicht-am-diff-bewerten/", "markdownUrl": "https://huecki.com/blog/ki-code-nicht-am-diff-bewerten.md", "publishedAt": "2026-05-28", "updatedAt": "2026-05-28", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "KI-Agenten", "Coding Agents", "Developer Workflow", "Agent Harness", "Evals" ], "summary": "Besseres KI-Coding entsteht nicht primär durch bessere Prompts, sondern durch den Harness um das Modell: explizite Contracts, getrennte Builder- und Reviewer-Rollen, Belege und eine Schleife, die Fehler in bessere Spezifikationen zurückführt.", "related": [], "sourceRefs": [] }, { "slug": "ki-first-softwarearchitektur", "title": "AI-first Architektur: schneller entscheiden, sauber bleiben", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-first-softwarearchitektur/", "markdownUrl": "https://huecki.com/blog/ki-first-softwarearchitektur.md", "publishedAt": "2026-04-29", "updatedAt": "2026-05-12", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "KI", "Softwarearchitektur", "GEO", "Engineering" ], "summary": "AI-first Architektur heißt nicht: Modell entscheidet. Es heißt: KI erzeugt Optionen, findet Risiken, verdichtet Kontext — das Team entscheidet und dokumentiert nachvollziehbar.", "related": [], "sourceRefs": [] }, { "slug": "ki-generierte-ui-braucht-playtester", "title": "Deine KI-generierte UI braucht einen Playtester, keinen Screenshot-Review", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-generierte-ui-braucht-playtester/", "markdownUrl": "https://huecki.com/blog/ki-generierte-ui-braucht-playtester.md", "publishedAt": "2026-05-28", "updatedAt": "2026-05-28", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "KI-Agenten", "Webwright", "Playwright", "UI Testing", "Coding Agents", "Developer Workflow", "Evals" ], "summary": "KI-generierte Interfaces sehen oft fertig aus, bevor sie sich korrekt verhalten. Eine GUI-Playtester-Loop schickt einen separaten Browser-Agenten in die App, protokolliert Interaktionen, speichert Screenshots und Logs, macht aus kaputten Flows reproduzierbare Bug Reports und rerunnt denselben Test nach dem Fix.", "related": [], "sourceRefs": [] }, { "slug": "ki-selbstcheck-findet-immer-etwas", "title": "Hör auf, KI zum kritischen Selbstcheck zu bitten", "language": "de", "canonicalUrl": "https://huecki.com/blog/ki-selbstcheck-findet-immer-etwas/", "markdownUrl": "https://huecki.com/blog/ki-selbstcheck-findet-immer-etwas.md", "publishedAt": "2026-05-21", "updatedAt": "2026-05-21", "topic": "KI-Agent Workflows", "agentTopicIds": [ "agent-security", "context-engineering", "agent-evals" ], "tags": [ "KI-Agenten", "Prompt Engineering", "Evals", "KI-Sicherheit", "Developer Workflow" ], "summary": "Offene Anweisungen wie „prüf das kritisch“ belohnen das Modell ungewollt dafür, Kritik zu produzieren. Die Lösung ist nicht weniger Review, sondern kalibriertes Review: klare Kriterien, PASS_NO_CHANGE, Evidenz pro Finding, Severity-Schwellen und ein kleines Änderungsbudget.", "related": [], "sourceRefs": [] }, { "slug": "llm-native-developer-operational-maturity", "title": "The LLM-native developer needs more than prompts", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/llm-native-developer-operational-maturity/", "markdownUrl": "https://huecki.com/en/blog/llm-native-developer-operational-maturity.md", "publishedAt": "2026-05-15", "updatedAt": "2026-05-15", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "LLM", "Software Architecture", "Agents", "Developer Workflow" ], "summary": "The next developer skill is not writing clever prompts. It is building the operating system around LLMs: data quality, model versioning, evals, guardrails, incident response, review UX, and repo instructions agents can actually follow.", "related": [], "sourceRefs": [] }, { "slug": "llm-native-entwickler-operational-maturity", "title": "LLM-native Entwickler brauchen mehr als gute Prompts", "language": "de", "canonicalUrl": "https://huecki.com/blog/llm-native-entwickler-operational-maturity/", "markdownUrl": "https://huecki.com/blog/llm-native-entwickler-operational-maturity.md", "publishedAt": "2026-05-15", "updatedAt": "2026-05-15", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "LLM", "Softwarearchitektur", "Agents", "Developer Workflow" ], "summary": "Die nächste Entwicklerfähigkeit ist nicht der cleverste Prompt. Es ist das Betriebssystem um LLMs herum: Datenqualität, Model-Versioning, Evals, Guardrails, Incident Response, Review-UX und Repo-Anweisungen, denen Agents wirklich folgen können.", "related": [], "sourceRefs": [] }, { "slug": "local-llm-agent-runtime-audit", "title": "Audit Local LLM Agents Like Runtimes", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/local-llm-agent-runtime-audit/", "markdownUrl": "https://huecki.com/en/blog/local-llm-agent-runtime-audit.md", "publishedAt": "2026-06-24", "updatedAt": "2026-06-24", "topic": "AI Agent Security", "agentTopicIds": [ "agent-security", "context-engineering", "agent-evals" ], "tags": [ "AI Security", "Local LLMs", "Agents", "Developer Workflow", "Runtime Security" ], "summary": "Local LLM agents can touch shells, files, browsers, credentials, memory, and messaging tools. Treat their runtime layer as source code worth auditing, then turn static findings into a manual review queue instead of automatic verdicts.", "related": [], "sourceRefs": [] }, { "slug": "measure-agentic-setups-skills", "title": "Don’t Benchmark the Model. Benchmark the Agent System.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/measure-agentic-setups-skills/", "markdownUrl": "https://huecki.com/en/blog/measure-agentic-setups-skills.md", "publishedAt": "2026-05-26", "updatedAt": "2026-05-26", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Agent Harness", "Evals", "Developer Workflow", "AI Engineering" ], "summary": "Agent evals should not only ask whether the final answer looked good. A useful benchmark measures the whole agent system: skill routing, tool policy, evidence, outcomes, hard-fail safety cases, regressions, cost, and production drift.", "related": [], "sourceRefs": [] }, { "slug": "messenger-domain-agents-ai-governance", "title": "Wie ich aus einer Messenger-App ein Betriebssystem für Domain Agents gebaut habe", "language": "de", "canonicalUrl": "https://huecki.com/blog/messenger-domain-agents-ai-governance/", "markdownUrl": "https://huecki.com/blog/messenger-domain-agents-ai-governance.md", "publishedAt": "2026-08-07", "updatedAt": "2026-08-07", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering" ], "tags": [ "AI Agents", "AI Governance", "Agent Workflows", "Agent Skills", "Automation", "Context Engineering" ], "summary": "Ich betreibe nicht für jede Aufgabe einen eigenen Bot. Mehrere getrennte Chat-Sessions greifen auf dieselbe agentische Runtime zu. Im zentralen Boba-DM entwerfe ich nach einem festen Factory-Playbook neue Domain-Profile aus Regeln, Memory, Skills, Tools und überprüfbaren Flows.", "related": [ { "slug": "agent-operating-contract-not-bigger-prompt", "relation": "Operating Contracts machen Ziele, Werkzeuge, Freigaben und Verifikation eines Agentenlaufs explizit." }, { "slug": "measure-agentic-setups-skills", "relation": "Ein Domain-Agent-System sollte als kompletter Lauf statt nur anhand seiner letzten Antwort bewertet werden." }, { "slug": "proactive-agent-notification-budget", "relation": "Proaktive Agenten brauchen Regeln dafür, wann eine Benachrichtigung wirklich hilfreich ist." } ], "sourceRefs": [] }, { "slug": "natural-language-agent-harnesses", "title": "Agent harnesses should be specs, not hidden glue code", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/natural-language-agent-harnesses/", "markdownUrl": "https://huecki.com/en/blog/natural-language-agent-harnesses.md", "publishedAt": "2026-05-24", "updatedAt": "2026-05-24", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "AI Agents", "Agent Harness", "Evals", "Developer Workflow" ], "summary": "Natural-Language Agent Harnesses give a useful name to an important shift: the agent policy should be an inspectable document that a runtime executes, not invisible glue hidden inside controller code.", "related": [], "sourceRefs": [] }, { "slug": "next-prompt-is-a-workflow", "title": "The Next Prompt Is Not a Prompt. It’s a Workflow.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/next-prompt-is-a-workflow/", "markdownUrl": "https://huecki.com/en/blog/next-prompt-is-a-workflow.md", "publishedAt": "2026-06-03", "updatedAt": "2026-06-03", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-security", "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Claude Code", "Agent Workflows", "Developer Workflow", "AI Engineering" ], "summary": "Dynamic workflows move agent work from one chat prompt into inspectable orchestration: phases, subagents, evidence, budget, permissions, adversarial review, and stop conditions. The point is not more agents. The point is better control.", "related": [], "sourceRefs": [] }, { "slug": "perfect-automated-ai-agent-eval-stack", "title": "The Perfect Automated AI Eval Stack Does Not Exist", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/perfect-automated-ai-agent-eval-stack/", "markdownUrl": "https://huecki.com/en/blog/perfect-automated-ai-agent-eval-stack.md", "publishedAt": "2026-07-15", "updatedAt": "2026-07-15", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Agents", "Evals", "LLM Observability", "Developer Workflow", "AI Engineering" ], "summary": "The reliable eval system is not one automated judge. It is a closed loop that combines portable traces, deterministic invariants, narrow semantic judges, versioned production failures, adversarial tests, and human calibration.", "related": [ { "slug": "agent-eval-too-short-trajectory", "relation": "Why final-answer scoring is insufficient for long-running agents." }, { "slug": "measure-agentic-setups-skills", "relation": "How to evaluate the complete agentic setup rather than the model alone." } ], "sourceRefs": [ { "title": "Do Automated Evals Work?", "url": "https://parlance-labs.com/blog/posts/auto-evals/", "type": "experiment" }, { "title": "Who Validates the Validators?", "url": "https://arxiv.org/abs/2404.12272", "type": "paper" }, { "title": "Judging the Judges: A Systematic Study of Position Bias in LLM-as-a-Judge", "url": "https://arxiv.org/abs/2406.07791", "type": "paper" }, { "title": "G-Eval", "url": "https://arxiv.org/abs/2303.16634", "type": "paper" }, { "title": "Arize Phoenix documentation", "url": "https://arize.com/docs/phoenix", "type": "documentation" }, { "title": "LangSmith evaluation concepts", "url": "https://docs.langchain.com/langsmith/evaluation-concepts", "type": "documentation" } ] }, { "slug": "prompting-2026-context-engineering", "title": "Prompting ist tot. Context zählt.", "language": "de", "canonicalUrl": "https://huecki.com/blog/prompting-2026-context-engineering/", "markdownUrl": "https://huecki.com/blog/prompting-2026-context-engineering.md", "publishedAt": "2026-05-12", "updatedAt": "2026-05-12", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "Prompt Engineering", "Context Engineering", "AI Agents", "KI-Workflows" ], "summary": "2026 geht es nicht mehr um den einen magischen Prompt. Der bessere Ansatz: Kontext wählen, Tools und Schemas definieren, Agent-Regeln setzen und mit Evals prüfen.", "related": [], "sourceRefs": [] }, { "slug": "prompting-2026-context-engineering-en", "title": "Prompting Is Dead. Context Wins.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/prompting-2026-context-engineering-en/", "markdownUrl": "https://huecki.com/en/blog/prompting-2026-context-engineering-en.md", "publishedAt": "2026-05-12", "updatedAt": "2026-05-12", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "Prompt Engineering", "Context Engineering", "AI Agents", "AI Workflows" ], "summary": "In 2026, good prompting is not about one magic sentence. The better approach is to curate context, define tools and schemas, set agent rules, and verify behavior with evals.", "related": [], "sourceRefs": [] }, { "slug": "prompting-2026-decomposition-skills-evals", "title": "Prompt Decomposition: So zerlegst du KI-Aufgaben richtig", "language": "de", "canonicalUrl": "https://huecki.com/blog/prompting-2026-decomposition-skills-evals/", "markdownUrl": "https://huecki.com/blog/prompting-2026-decomposition-skills-evals.md", "publishedAt": "2026-05-18", "updatedAt": "2026-05-18", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "Prompt Engineering", "Decomposition", "AI Agents", "Skills", "Developer Workflow" ], "summary": "Nach Context Engineering kommt Decomposition: Entwickler sollten nicht alles in einen Prompt stopfen, sondern Aufgaben in direkte Prompts, Subtasks, Pipelines, Agent-Loops oder Skills zerlegen.", "related": [], "sourceRefs": [] }, { "slug": "prompting-2026-decomposition-skills-evals-en", "title": "Prompt Decomposition: How to Break Down AI Tasks Properly", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/prompting-2026-decomposition-skills-evals-en/", "markdownUrl": "https://huecki.com/en/blog/prompting-2026-decomposition-skills-evals-en.md", "publishedAt": "2026-05-18", "updatedAt": "2026-05-18", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "Prompt Engineering", "Decomposition", "AI Agents", "Skills", "Developer Workflow" ], "summary": "After context engineering comes decomposition: developers should stop putting everything into one prompt and instead split tasks into direct prompts, subtasks, pipelines, agent loops, or skills.", "related": [], "sourceRefs": [] }, { "slug": "rag-2026-contextual-hybrid-rag", "title": "RAG 2026: Vergiss GraphRAG als Default", "language": "de", "canonicalUrl": "https://huecki.com/blog/rag-2026-contextual-hybrid-rag/", "markdownUrl": "https://huecki.com/blog/rag-2026-contextual-hybrid-rag.md", "publishedAt": "2026-06-09", "updatedAt": "2026-06-09", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "RAG", "Context Engineering", "AI Engineering", "LLM", "Evals" ], "summary": "RAG wird 2026 nicht dadurch besser, dass man GraphRAG auf alles wirft. Der robuste Default ist Contextual Hybrid RAG: saubere Ingestion, BM25 plus Embeddings, Reranking, Quellenpflicht und harte Evals.", "related": [], "sourceRefs": [] }, { "slug": "should-ai-agent-remember-what-it-learned", "title": "Your AI Agent Learned Something. Should It Be Allowed to Remember It?", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/should-ai-agent-remember-what-it-learned/", "markdownUrl": "https://huecki.com/en/blog/should-ai-agent-remember-what-it-learned.md", "publishedAt": "2026-07-17", "updatedAt": "2026-07-17", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-evals", "agent-security", "context-engineering" ], "tags": [ "AI Agents", "Agent Memory", "Self-Improvement", "Agent Evals", "AI Security" ], "summary": "An agent that writes a lesson into memory, a skill, a prompt, or its own code is deploying behavior into future runs. This guide shows how to put persistent changes through evidence, eval, approval, expiry, and rollback gates.", "related": [ { "slug": "agent-is-not-the-product-fitness-function-is", "relation": "Explains the external fitness function and promotion gate around self-improving agents." }, { "slug": "ai-self-reflection-defends-first-answer", "relation": "Separates within-run correction from durable changes across future runs." }, { "slug": "hermes-self-improving-agent", "relation": "Shows a practical skill-improvement loop that needs persistence boundaries." } ], "sourceRefs": [ { "title": "Self-Improvements in Modern Agentic Systems: A Survey", "url": "https://arxiv.org/abs/2607.13104", "type": "paper" }, { "title": "A Self-Improving Coding Agent", "url": "https://arxiv.org/abs/2504.15228", "type": "paper" }, { "title": "Reflexion: Language Agents with Verbal Reinforcement Learning", "url": "https://arxiv.org/abs/2303.11366", "type": "paper" }, { "title": "Voyager: An Open-Ended Embodied Agent with Large Language Models", "url": "https://arxiv.org/abs/2305.16291", "type": "paper" }, { "title": "From Untrusted Input to Trusted Memory", "url": "https://arxiv.org/abs/2606.04329", "type": "paper" }, { "title": "When Agents Remember Too Much", "url": "https://arxiv.org/abs/2607.06595", "type": "paper" } ] }, { "slug": "simple-graph-rag-hallucinations", "title": "Simple Graph-RAG halbiert Halluzinationen – ohne komplexes Knowledge Graph Schema", "language": "de", "canonicalUrl": "https://huecki.com/blog/simple-graph-rag-hallucinations/", "markdownUrl": "https://huecki.com/blog/simple-graph-rag-hallucinations.md", "publishedAt": "2026-07-02", "updatedAt": "2026-07-02", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-evals", "llm-native-engineering" ], "tags": [ "RAG", "GraphRAG", "Hallucinations", "Knowledge Graph", "AI Engineering", "QA" ], "summary": "Vector-RAG + einfacher Document-Graph halbiert Halluzinationen auf komplexen QA-Aufgaben im MoNaCo-Benchmark. Die Erkenntnis: Man braucht kein komplexes Knowledge Graph Schema – nur strukturierte Navigation über Dokumente.", "related": [], "sourceRefs": [] }, { "slug": "single-turn-evals-dont-teach-agent-enough", "title": "Single-Turn Evals Don’t Teach Your Agent Enough", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/single-turn-evals-dont-teach-agent-enough/", "markdownUrl": "https://huecki.com/en/blog/single-turn-evals-dont-teach-agent-enough.md", "publishedAt": "2026-08-23", "updatedAt": "2026-08-23", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Agent Skills", "Agent Evals", "AI Engineering", "Automation" ], "summary": "Single-turn evals expose the first obvious skill gap, then stop teaching the system. Use multi-turn failure replay, repairable attribution, bounded edits, and a separate governance check instead.", "related": [ { "slug": "stop-losing-agent-workflows", "relation": "FlowEvo field note on turning successful traces into reusable skills." }, { "slug": "measure-agentic-setups-skills", "relation": "Broader guide to evaluating skills, tools, traces, and agent systems." }, { "slug": "your-agent-harness-needs-a-release-process", "relation": "Release-process frame for shipping prompt, tool, memory, and harness changes." } ], "sourceRefs": [ { "title": "SkillEvo: Self-Renewing Evolution Gradients from Multi-Turn Interaction Feedback", "url": "https://arxiv.org/html/2608.13120v1", "type": "paper" }, { "title": "SkillEvo arXiv abstract", "url": "https://arxiv.org/abs/2608.13120", "type": "paper" }, { "title": "Hugging Face paper page: SkillEvo", "url": "https://huggingface.co/papers/2608.13120", "type": "paper-index" } ] }, { "slug": "spec-driven-context-resets-for-coding-agents", "title": "Spec-Driven Context Resets for Coding Agents", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/spec-driven-context-resets-for-coding-agents/", "markdownUrl": "https://huecki.com/en/blog/spec-driven-context-resets-for-coding-agents.md", "publishedAt": "2026-05-23", "updatedAt": "2026-05-23", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "llm-native-engineering" ], "tags": [ "AI Engineering", "Coding Agents", "Context Engineering", "Spec-Driven Development", "Developer Workflow" ], "summary": "Long agent chats rot. A better pattern is to move decisions into small spec files, clear context between layers, and let each coding-agent session read only the artifact it needs.", "related": [], "sourceRefs": [] }, { "slug": "spec-driven-context-resets-fuer-coding-agenten", "title": "Spec-Driven Context Resets für Coding-Agenten", "language": "de", "canonicalUrl": "https://huecki.com/blog/spec-driven-context-resets-fuer-coding-agenten/", "markdownUrl": "https://huecki.com/blog/spec-driven-context-resets-fuer-coding-agenten.md", "publishedAt": "2026-05-23", "updatedAt": "2026-05-23", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "llm-native-engineering" ], "tags": [ "AI Engineering", "Coding Agents", "Context Engineering", "Spec-Driven Development", "Developer Workflow" ], "summary": "Lange Agenten-Chats verrotten. Besser ist es, Entscheidungen in kleine Spec-Dateien zu verschieben, zwischen den Ebenen bewusst den Kontext zu resetten und jede Coding-Agent-Session nur das lesen zu lassen, was sie wirklich braucht.", "related": [], "sourceRefs": [] }, { "slug": "sprachnachrichten-agenten-interface", "title": "Sprachnachrichten sind das beste Interface für kleine Agentenjobs", "language": "de", "canonicalUrl": "https://huecki.com/blog/sprachnachrichten-agenten-interface/", "markdownUrl": "https://huecki.com/blog/sprachnachrichten-agenten-interface.md", "publishedAt": "2026-05-15", "updatedAt": "2026-05-15", "topic": "Personal AI Workflows", "agentTopicIds": [ "context-engineering" ], "tags": [ "AI Agents", "Voice Interface", "Automation", "Open Source", "Personal AI" ], "summary": "Voice ist nicht gut für alles. Aber für kleine Agentenjobs ist es brutal praktisch: unterwegs eine Aufgabe diktieren, lokal transkribieren, vom bestehenden Agenten ausführen lassen und nur eine kurze Antwort zurückbekommen.", "related": [], "sourceRefs": [] }, { "slug": "stop-asking-which-coding-model-is-best", "title": "Stop Asking Which Coding Model Is Best", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-asking-which-coding-model-is-best/", "markdownUrl": "https://huecki.com/en/blog/stop-asking-which-coding-model-is-best.md", "publishedAt": "2026-07-10", "updatedAt": "2026-07-10", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals" ], "tags": [ "AI Agents", "Agent Harness", "Coding Agents", "Agent Evals", "AI Engineering" ], "summary": "The useful question is moving from which model is best to what your agent harness can change, measure, persist, and roll back.", "related": [ { "slug": "agents-dont-need-longer-prompts-they-need-harnesses", "relation": "Earlier framing of the harness as the runtime around the model." }, { "slug": "measure-agentic-setups-skills", "relation": "How to evaluate skills, traces, tools, and the whole agent run." }, { "slug": "agent-is-not-the-product-fitness-function-is", "relation": "Why external evaluation systems matter more than self-improvement vibes." } ], "sourceRefs": [ { "title": "TTHE: Test-Time Harness Evolution", "url": "https://arxiv.org/abs/2607.08124", "type": "paper" }, { "title": "SkillOpt: Executive Strategy for Self-Evolving Agent Skills", "url": "https://arxiv.org/abs/2605.23904", "type": "paper" }, { "title": "microsoft/SkillOpt", "url": "https://github.com/microsoft/SkillOpt", "type": "repository" }, { "title": "SkillOpt: Agent skills as trainable parameters", "url": "https://www.microsoft.com/en-us/research/blog/skillopt-agent-skills-as-trainable-parameters/", "type": "official" }, { "title": "The Harness Effect: How Orchestration Design Sets the Token Economics of Enterprise Agentic AI", "url": "https://arxiv.org/abs/2607.06906", "type": "paper" }, { "title": "Evolving how LLMs are measured for Android: the next era of Android Bench", "url": "https://android-developers.googleblog.com/2026/07/android-bench-llm-measurement.html", "type": "official" }, { "title": "android-bench/android-bench", "url": "https://github.com/android-bench/android-bench", "type": "repository" } ] }, { "slug": "stop-judging-ai-code-by-the-diff", "title": "Stop Judging AI Code by the Diff", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-judging-ai-code-by-the-diff/", "markdownUrl": "https://huecki.com/en/blog/stop-judging-ai-code-by-the-diff.md", "publishedAt": "2026-05-28", "updatedAt": "2026-05-28", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "AI Agents", "Coding Agents", "Developer Workflow", "Agent Harness", "Evals" ], "summary": "Better AI coding is not mainly about better prompts. It is about the harness around the model: explicit contracts, separate builder and reviewer roles, evidence requirements, and a loop that turns failures into better specifications.", "related": [], "sourceRefs": [] }, { "slug": "stop-losing-agent-workflows", "title": "Stop Losing the Workflow Your Agent Just Discovered", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-losing-agent-workflows/", "markdownUrl": "https://huecki.com/en/blog/stop-losing-agent-workflows.md", "publishedAt": "2026-08-23", "updatedAt": "2026-08-23", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering" ], "tags": [ "AI Agents", "Agent Skills", "Developer Workflow", "AI Engineering", "Automation" ], "summary": "Successful agent traces should not disappear when the session ends. Extract the reusable part, wrap it in an interface, replay it, and only then admit it into a skill bank.", "related": [ { "slug": "failed-agent-runs-should-become-skills", "relation": "Earlier draft on mining failed traces into local skills" }, { "slug": "agent-recipes-not-bigger-prompts", "relation": "Related field note on durable agent recipes" } ], "sourceRefs": [ { "title": "FlowEvo: Self-Evolving Agents through the Co-Evolution of Workflows and Executable Skills", "url": "https://arxiv.org/abs/2607.21596", "type": "paper" }, { "title": "DEFENSE-SEU/FlowEvo", "url": "https://github.com/DEFENSE-SEU/FlowEvo", "type": "code" } ] }, { "slug": "stop-prompting-coding-agent-loop-spec", "title": "Stop Prompting Your Coding Agent. Give It a Loop.", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/stop-prompting-coding-agent-loop-spec/", "markdownUrl": "https://huecki.com/en/blog/stop-prompting-coding-agent-loop-spec.md", "publishedAt": "2026-07-02", "updatedAt": "2026-07-02", "topic": "AI Agent Workflows", "agentTopicIds": [ "context-engineering", "llm-native-engineering" ], "tags": [ "AI Agents", "Coding Agents", "Agent Skills", "Agent Memory", "Developer Workflow", "Verification" ], "summary": "The useful upgrade from prompt engineering is not a longer instruction block. It is a reusable loop spec: trigger, goal, allowed tools, verifier, terminal states, and memory rules. That is how repeated coding-agent work becomes operational instead of conversational.", "related": [ { "slug": "agents-dont-need-longer-prompts-they-need-harnesses", "relation": "The harness is the runtime; the loop spec is the reusable operating procedure you hand to it." }, { "slug": "failed-agent-runs-should-become-skills", "relation": "Failed runs provide evidence for improving the reusable skill inside a loop." }, { "slug": "agent-memory-tests-measure-wrong-thing", "relation": "Loop memory should be tested for update, scope, and false-premise behavior." } ], "sourceRefs": [ { "title": "Stop Hand-Holding Your Coding Agent: Engineering the Loops that Replace Step-by-Step Prompting", "url": "https://arxiv.org/abs/2607.00038", "type": "paper" }, { "title": "SkillOpt: Executive Strategy for Self-Evolving Agent Skills", "url": "https://arxiv.org/abs/2605.23904", "type": "paper" }, { "title": "MemSyco-Bench: Benchmarking Sycophancy in Agent Memory", "url": "https://arxiv.org/abs/2607.01071", "type": "paper" }, { "title": "Huecki AI radar, 2026-07-02", "url": "https://huecki.com", "type": "internal-radar" } ] }, { "slug": "voice-notes-agent-interface", "title": "Voice notes are the best interface for small agent jobs", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/voice-notes-agent-interface/", "markdownUrl": "https://huecki.com/en/blog/voice-notes-agent-interface.md", "publishedAt": "2026-05-15", "updatedAt": "2026-05-15", "topic": "Personal AI Workflows", "agentTopicIds": [ "context-engineering" ], "tags": [ "AI Agents", "Voice Interface", "Automation", "Open Source", "Personal AI" ], "summary": "Voice is not good for everything. But for small agent jobs it is brutally useful: dictate a task while moving, transcribe it locally, let your existing agent handle it, and get only a short answer back.", "related": [], "sourceRefs": [] }, { "slug": "von-prompt-klappern-zu-prompt-debugging", "title": "Von 'Prompt-Klappern' zu Prompt-Debugging", "language": "de", "canonicalUrl": "https://huecki.com/blog/von-prompt-klappern-zu-prompt-debugging/", "markdownUrl": "https://huecki.com/blog/von-prompt-klappern-zu-prompt-debugging.md", "publishedAt": "2026-07-01", "updatedAt": "2026-07-01", "topic": "AI-first Engineering", "agentTopicIds": [ "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Engineering", "Prompt Engineering", "LLM Agents", "RAG", "Evals", "KDD 2026" ], "summary": "Prompt-Optimierung ist kein Suchproblem mehr — es ist ein Debugging-Problem. Contrastive Reflection liefert das Framework, um Fehler gezielt zu fangen, statt blind zu raten.", "related": [], "sourceRefs": [] }, { "slug": "your-agent-harness-needs-a-release-process", "title": "Your Agent Harness Needs a Release Process", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/your-agent-harness-needs-a-release-process/", "markdownUrl": "https://huecki.com/en/blog/your-agent-harness-needs-a-release-process.md", "publishedAt": "2026-08-11", "updatedAt": "2026-08-11", "topic": "AI Agent Workflows", "agentTopicIds": [ "agent-harnesses", "context-engineering", "agent-evals", "llm-native-engineering" ], "tags": [ "AI Agents", "Agent Harnesses", "Agent Evals", "Observability", "AI Governance" ], "summary": "A practical field note on operating agent-harness changes like product releases: start from a trace-backed failure, change one bounded component, evaluate repeated trials and private holdouts, then promote through review with a rollback path.", "related": [ { "slug": "harness-engineering-field-guide", "relation": "Maps the complete engineering surface around the model; this note focuses on how changes to that surface should ship." }, { "slug": "agent-harness-needs-a-behavior-map", "relation": "A behavior map helps locate which harness component actually owns the observed failure." }, { "slug": "your-ai-agent-needs-a-replay-button", "relation": "Replayable traces turn production failures into regression cases for the next harness release." } ], "sourceRefs": [ { "title": "Ouroboros: A Self-Developing Frontier Coding Agent with Reviewed Core Evolution", "url": "https://arxiv.org/abs/2608.08311", "type": "paper" }, { "title": "Evo-Bench: Can Language Models Improve Agent Harness?", "url": "https://arxiv.org/abs/2608.09096", "type": "paper" }, { "title": "A2E: An End-to-End Agent Auditing Engine", "url": "https://arxiv.org/abs/2608.07346", "type": "paper" }, { "title": "Rethinking the Evaluation of Harness Evolution for Agents", "url": "https://arxiv.org/abs/2607.12227", "type": "paper" }, { "title": "Demystifying evals for AI agents", "url": "https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents", "type": "official" } ] }, { "slug": "your-onboarding-is-why-your-team-is-vibe-coding", "title": "Your Onboarding Is Why Your Team Is Vibe Coding", "language": "en", "canonicalUrl": "https://huecki.com/en/blog/your-onboarding-is-why-your-team-is-vibe-coding/", "markdownUrl": "https://huecki.com/en/blog/your-onboarding-is-why-your-team-is-vibe-coding.md", "publishedAt": "2026-05-20", "updatedAt": "2026-05-20", "topic": "AI-first Engineering", "agentTopicIds": [ "agent-harnesses", "context-engineering", "llm-native-engineering" ], "tags": [ "AI Engineering", "Developer Onboarding", "Vibe Coding", "Developer Experience", "Internal Tools", "Agent Harness" ], "summary": "Teams do not usually start vibe coding because developers became careless. They start because onboarding is broken: docs are stale, harnesses are undocumented, system knowledge lives in people’s heads, and AI turns missing context into plausible code and Markdown.", "related": [], "sourceRefs": [] } ], "tools": [ { "slug": "agent-save-file-starter", "title": "Agent Save File Starter", "language": "de", "canonicalUrl": "https://huecki.com/tools/agent-save-file-starter/", "markdownUrl": "https://huecki.com/tools/agent-save-file-starter.md", "publishedAt": "2026-06-16", "updatedAt": "2026-06-16", "category": "Template", "status": "Experimental", "agentTopicIds": [ "agent-harnesses", "context-engineering" ], "tags": [ "Coding Agents", "Agent Memory", "Context Engineering", "AGENTS.md" ], "summary": "Ein leichtgewichtiges Projektgerüst für Agent-Kontextdateien. Es trennt harte Regeln, sichere Commands, gültige Entscheidungen und alte Hinweise, damit Agents nicht aus Chat-Müll arbeiten.", "related": [], "sourceRefs": [] }, { "slug": "agent-save-file-starter-en", "title": "Agent Save File Starter", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/agent-save-file-starter-en/", "markdownUrl": "https://huecki.com/en/tools/agent-save-file-starter-en.md", "publishedAt": "2026-06-16", "updatedAt": "2026-06-16", "category": "Template", "status": "Experimental", "agentTopicIds": [ "agent-harnesses", "context-engineering" ], "tags": [ "Coding Agents", "Agent Memory", "Context Engineering", "AGENTS.md" ], "summary": "A lightweight project structure for agent context files. It separates hard rules, safe commands, active decisions, and archived notes so agents do not work from chat sludge.", "related": [], "sourceRefs": [] }, { "slug": "architecture-diagram-mcp", "title": "Architecture Diagram MCP", "language": "de", "canonicalUrl": "https://huecki.com/tools/architecture-diagram-mcp/", "markdownUrl": "https://huecki.com/tools/architecture-diagram-mcp.md", "publishedAt": "2026-06-22", "updatedAt": "2026-06-22", "category": "MCP", "status": "Experimental", "agentTopicIds": [ "llm-native-engineering" ], "tags": [ "MCP", "Architecture", "Diagrams", "Agentic Platform" ], "summary": "Architecture Diagram MCP erzeugt browserfähige Architekturartefakte aus Agent-Briefs. Es ist für Plattformkarten, Stakeholder-Erklärungen und reproduzierbare Diagramm-Outputs gebaut.", "related": [], "sourceRefs": [] }, { "slug": "architecture-diagram-mcp-en", "title": "Architecture Diagram MCP", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/architecture-diagram-mcp-en/", "markdownUrl": "https://huecki.com/en/tools/architecture-diagram-mcp-en.md", "publishedAt": "2026-06-22", "updatedAt": "2026-06-22", "category": "MCP", "status": "Experimental", "agentTopicIds": [ "llm-native-engineering" ], "tags": [ "MCP", "Architecture", "Diagrams", "Agentic Platform" ], "summary": "Architecture Diagram MCP creates browser-ready architecture artifacts from agent briefs. It is built for platform maps, stakeholder explainers, and reproducible diagram outputs.", "related": [], "sourceRefs": [] }, { "slug": "company-researcher", "title": "Company Researcher", "language": "de", "canonicalUrl": "https://huecki.com/tools/company-researcher/", "markdownUrl": "https://huecki.com/tools/company-researcher.md", "publishedAt": "2026-06-22", "updatedAt": "2026-06-22", "category": "Tool", "status": "Experimental", "agentTopicIds": [ "tool" ], "tags": [ "Company Research", "Bundesanzeiger", "OpenRouter", "Next.js" ], "summary": "Company Researcher ist die produktive Oberfläche für schnelle Firmenrecherche: Unternehmensname eingeben, Research-Flow starten, Finanzdaten einsammeln und einen lesbaren Report erzeugen.", "related": [], "sourceRefs": [] }, { "slug": "company-researcher-en", "title": "Company Researcher", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/company-researcher-en/", "markdownUrl": "https://huecki.com/en/tools/company-researcher-en.md", "publishedAt": "2026-06-22", "updatedAt": "2026-06-22", "category": "Tool", "status": "Experimental", "agentTopicIds": [ "tool" ], "tags": [ "Company Research", "Bundesanzeiger", "OpenRouter", "Next.js" ], "summary": "Company Researcher is the production surface for quick company research: enter a company name, run the research flow, collect financial signals, and produce a readable report.", "related": [], "sourceRefs": [] }, { "slug": "deepresearch-skill", "title": "DeepResearch Skill", "language": "de", "canonicalUrl": "https://huecki.com/tools/deepresearch-skill/", "markdownUrl": "https://huecki.com/tools/deepresearch-skill.md", "publishedAt": "2026-06-16", "updatedAt": "2026-06-16", "category": "Skill", "status": "Stable", "agentTopicIds": [ "context-engineering" ], "tags": [ "Research", "Agent Skills", "Evidence", "Sources" ], "summary": "DeepResearch zwingt Agenten aus dem schnellen Suchmaschinen-Summary-Modus heraus. Der Skill schreibt Recherche-Artefakte, trennt Primär- von Sekundärquellen und verlangt URL-gestützte Evidenz vor der Synthese.", "related": [], "sourceRefs": [] }, { "slug": "deepresearch-skill-en", "title": "DeepResearch Skill", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/deepresearch-skill-en/", "markdownUrl": "https://huecki.com/en/tools/deepresearch-skill-en.md", "publishedAt": "2026-06-16", "updatedAt": "2026-06-16", "category": "Skill", "status": "Stable", "agentTopicIds": [ "context-engineering" ], "tags": [ "Research", "Agent Skills", "Evidence", "Sources" ], "summary": "DeepResearch pushes agents out of quick search-summary mode. The skill writes research artifacts, separates primary from secondary sources, and requires URL-backed evidence before synthesis.", "related": [], "sourceRefs": [] }, { "slug": "is-this-agentic-change-good", "title": "Is This Agentic Change Good?", "language": "de", "canonicalUrl": "https://huecki.com/tools/is-this-agentic-change-good/", "markdownUrl": "https://huecki.com/tools/is-this-agentic-change-good.md", "publishedAt": "2026-06-16", "updatedAt": "2026-06-16", "category": "Skill", "status": "Experimental", "agentTopicIds": [ "agent-harnesses", "agent-security", "context-engineering", "agent-evals" ], "tags": [ "Agent Skills", "Review", "AGENTS.md", "Safety" ], "summary": "Dieser Skill prüft Branches, die zukünftiges Agent-Verhalten verändern. Er sammelt lokale Git-Evidenz, bewertet Skill-Kollisionen, Kontextbudget, Sicherheitsrisiken und Runtime-Sichtbarkeit, bevor ein Merge empfohlen wird.", "related": [], "sourceRefs": [] }, { "slug": "is-this-agentic-change-good-en", "title": "Is This Agentic Change Good?", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/is-this-agentic-change-good-en/", "markdownUrl": "https://huecki.com/en/tools/is-this-agentic-change-good-en.md", "publishedAt": "2026-06-16", "updatedAt": "2026-06-16", "category": "Skill", "status": "Experimental", "agentTopicIds": [ "agent-harnesses", "agent-security", "context-engineering", "agent-evals" ], "tags": [ "Agent Skills", "Review", "AGENTS.md", "Safety" ], "summary": "This skill reviews branches that change future agent behavior. It collects local git evidence and judges skill collisions, context budget, safety risks, and runtime visibility before recommending a merge.", "related": [], "sourceRefs": [] }, { "slug": "presentation-maker-mcp", "title": "Presentation Maker MCP", "language": "de", "canonicalUrl": "https://huecki.com/tools/presentation-maker-mcp/", "markdownUrl": "https://huecki.com/tools/presentation-maker-mcp.md", "publishedAt": "2026-06-22", "updatedAt": "2026-06-22", "category": "MCP", "status": "Experimental", "agentTopicIds": [ "mcp" ], "tags": [ "MCP", "Presentations", "HTML Decks", "Agentic Platform" ], "summary": "Presentation Maker MCP baut browserfähige Decks aus Briefings. Es verbindet klare Slide-Struktur, Speaker Notes, Tastaturnavigation und exportierbare HTML-Artefakte.", "related": [], "sourceRefs": [] }, { "slug": "presentation-maker-mcp-en", "title": "Presentation Maker MCP", "language": "en", "canonicalUrl": "https://huecki.com/en/tools/presentation-maker-mcp-en/", "markdownUrl": "https://huecki.com/en/tools/presentation-maker-mcp-en.md", "publishedAt": "2026-06-22", "updatedAt": "2026-06-22", "category": "MCP", "status": "Experimental", "agentTopicIds": [ "mcp" ], "tags": [ "MCP", "Presentations", "HTML Decks", "Agentic Platform" ], "summary": "Presentation Maker MCP builds browser-ready decks from briefs. It combines slide structure, speaker notes, keyboard navigation, and exportable HTML artifacts.", "related": [], "sourceRefs": [] } ] }