The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands for repository inventory (find) and measurement. It uses a bundled Python script (scripts/agents_md_check.py) that may invoke the just CLI tool (just --summary) to retrieve a list of recipes from a project's justfile. This is used solely to verify that commands documented in the instruction files are valid and not stale.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it is designed to read and process the contents of repository instruction files (AGENTS.md, CLAUDE.md, etc.), which could theoretically contain malicious instructions. However, the skill incorporates strong guardrails to mitigate this risk.
Ingestion points: Reads repository-scoped instruction files (AGENTS.md, CLAUDE.md, GEMINI.md, .github/copilot-instructions.md) and project justfiles via a measurement script and shell commands.
Boundary markers: The instructions explicitly require the agent to present proposed changes as a diff and obtain manual user approval before writing any files.
Capability inventory: Includes file system read/write access and shell command execution (find, python3, just).
Sanitization: The skill's primary function is to audit these files against a provided checklist and the actual repository state to remove anti-patterns, secrets, and stale commands.

agents-md