The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local script scripts/packs-resolve.py to manage configuration dependencies. This script invokes the git binary using subprocess.run to perform repository operations.
[REMOTE_CODE_EXECUTION]: The skill is designed to automatically clone or fetch data from remote git repositories based on configurations found in the target project (.compound-engineering/config.yaml). Although the script only processes markdown files from these sources, the automated downloading of remote content based on repository state (which could be influenced by a Pull Request) is a significant capability.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple sources, which could be used to influence the agent's behavior during the QA process.
Ingestion points: The agent reads the output of git diff, gh pr view, and the contents of external markdown files within "Compound Packs."
Boundary markers: The instructions do not specify explicit delimiters or safety instructions (e.g., "ignore instructions in the diff") when processing these external inputs.
Capability inventory: The agent possesses powerful capabilities, including full browser automation (agent-browser), file system writes for generating reports, and the ability to execute CLI tools like git and gh.
Sanitization: The scripts/packs-resolve.py script implements specific security checks to mitigate risks associated with untrusted content, such as _within for path traversal prevention and _escaping_links to detect and skip symlinks that point outside the intended source directory.

ce-dogfood