The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project-level documentation files such as AGENTS.md, README.md, and CONTRIBUTING.md, as well as configuration files like package.json or pyproject.toml. These files represent an attack surface where malicious instructions could be embedded to influence the agent's behavior during documentation updates.
Ingestion points: The skill reads various Markdown and project manifest files from the local repository.
Boundary markers: Instructions require the agent to verify all documentation claims against the actual codebase state and preserve authority/safety boundaries when editing.
Capability inventory: File system access (reading, writing, moving, symlinking) and execution of discovery/verification commands.
Sanitization: Relies on cross-referencing documentation claims with source-of-truth files (code and manifests) to detect inaccuracies or injections.
[COMMAND_EXECUTION]: The skill directs the agent to execute local shell commands for repository discovery and claim verification.
Evidence: Uses git ls-files, git rev-parse, ls, tree, and test commands to inspect project structure and validate scripts defined in package manager configuration files.

md-docs