The Agent Skills Directory

[SAFE]: The skill operates as intended for planning and documentation within a development environment without engaging in suspicious activities.
[COMMAND_EXECUTION]: The skill executes a local script (scripts/metadata.py) to retrieve standard Git repository information (commit, branch, and origin URL). Evidence: The script uses subprocess.run with list-based arguments to call git remote, git rev-parse, and git branch. This information is used solely to populate metadata in the generated implementation plan.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and trust external research documents and codebase files, which is an inherent attack surface for this category of tool. 1. Ingestion points: User-provided file paths and documents located in docs/agents/research/. 2. Boundary markers: The instructions direct the agent to read files completely but do not define specific delimiters or 'ignore instructions' wrappers for the ingested content. 3. Capability inventory: The skill possesses file read/write capabilities and the ability to execute a local metadata script. 4. Sanitization: The skill process includes a mandatory review step where the agent presents the draft plan to the user for approval before finalization.

rpi-plan