The Agent Skills Directory

[SAFE]: The skill provides structured guidelines for repository documentation aimed at improving AI agent reliability and maintainability.- [COMMAND_EXECUTION]: Instructs the agent to run standard development toolchain commands (e.g., cargo test, clippy, doc, bun run) which are appropriate for its stated purpose of managing software projects and verifying documentation consistency.- [PROMPT_INJECTION]: The skill defines a framework for agents to ingest and act upon project-specific documentation (AGENTS.md, change-impact tables). While this creates a surface for indirect prompt injection if project files are manipulated, the skill itself provides beneficial structure and boundary markers.
Ingestion points: The agent reads project-level AGENTS.md, CLAUDE.md, and documentation files within the docs/ hierarchy.
Boundary markers: The architecture establishes AGENTS.md as the authoritative contract and entry point, using "Critical invariants" to define hard rules for agent behavior.
Capability inventory: The agent has the capability to modify project documentation and source code, and to execute local build/test commands via cargo and bun.
Sanitization: No explicit sanitization of documentation content is specified beyond the project's standard verification and linting commands.

agentic-docs-architecture