The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands for repository management and discovery, specifically git rev-parse --show-toplevel for identifying the repository root, fd for file discovery, and ln -sf AGENTS.md CLAUDE.md to maintain compatibility symlinks within the repository.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection risks as it processes repository data to generate documentation. 1. Ingestion points: Reads repository content from README.md, AGENTS.md, and project manifests such as package.json or Cargo.toml. 2. Boundary markers: The instructions lack specified delimiters or isolation mechanisms to prevent instructions within those files from influencing the agent's behavior. 3. Capability inventory: The agent is authorized to write Markdown files and manage filesystem symlinks within the repository scope. 4. Sanitization: The skill does not perform explicit sanitization or validation of the ingested repository text.
[SAFE]: The skill implements a 'Repository Guard Rail' shell script that performs active checks to prevent execution at the filesystem root, the user's home directory, or within specific agent configuration directories (e.g., .agents, .codex), which effectively mitigates accidental or malicious system-wide writes.

agents-context-management