The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes git commit history to generate release notes. Commit messages are external, untrusted data that could contain malicious instructions designed to influence the agent's subsequent actions during release note validation or publication.
Ingestion points: scripts/draft_release_notes.py reads commit subjects via git log.
Boundary markers: The generated release notes do not use delimiters or warnings to isolate the commit messages from the agent's instructions.
Capability inventory: The script can write to arbitrary file paths provided via the --out argument, and the agent is instructed to publish the resulting release package.
Sanitization: No sanitization, escaping, or validation is performed on the commit messages before they are interpolated into the release notes.
[COMMAND_EXECUTION]: The skill executes the git binary using the subprocess module to retrieve repository history.
Evidence: The run_git function in scripts/draft_release_notes.py uses subprocess.run with a list of arguments to execute git log. User-provided references are interpolated into the command arguments.

github-release-management