Agent Docs

**Quick judgment:** A strong, opinionated skill for producing AI-agent-optimized documentation that follows current best practices (Hybrid Context Hierarchy, AGENTS.md, llms.txt). Well-suited for teams serious about agent reliability and RAG quality, with an emphasis on structure and constraints over prose. **Strengths & benefits** - Focuses on *inline constitutional docs* (AGENTS.md) to address meta-cognitive failures and avoid over-reliance on retrieval. - Encourages compressed, high-SNR indices (file paths, signatures, negative constraints) that are easier for agents and RAG systems to use. - Explicitly designs around known LLM weaknesses (chunking behavior, primacy/recency, lost-in-the-middle). - Integrates security awareness: no secrets inline, clear separation between trusted inline docs and external sources. **Key risks / limitations** - Relies on correct *human adoption*: if teams don’t actually maintain AGENTS.md / llms.txt as prescribed, benefits drop quickly. - Opinionated structure may not align with legacy doc sets; migration can be non-trivial for large codebases. - No direct evidence here of automated validation or linting; quality depends on how carefully the skill is used. **Recommended scenarios** - You are publishing OpenClaw skills, tools, or APIs and want agents to use them correctly and safely with minimal trial-and-error. - You maintain a codebase where agents assist with coding, debugging, or operations and need robust AGENTS.md / llms.txt. - You are designing or overhauling RAG/documentation for frameworks, SDKs, or complex backends and want to maximize agent pass rates while minimizing token usage.