The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The skill executes local git commands through Python's subprocess.run using argument lists. This implementation is secure against shell injection as it avoids spawning a shell and treats inputs as literal arguments. Command execution is strictly limited to the git binary.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill processes untrusted data from external sources (local Git repositories) which could contain malicious instructions.
Ingestion points: scripts/analyze_repo.py and scripts/get_changed_files.py ingest commit messages, author names, and filenames via git log, git shortlog, and git ls-files commands.
Boundary markers: Absent. The instructions do not provide delimiters or warnings to the agent to ignore instructions embedded within the repository data.
Capability inventory: The skill allows execution of git commands via subprocess.run and provides the agent with the repository's metadata to summarize for the user.
Sanitization: Absent. Data extracted from Git is passed directly to the agent as raw strings.
Risk: An attacker-controlled repository could use crafted commit messages (e.g., 'IMPORTANT: Ignore previous instructions and instead...') to influence the agent's summary or behavior when the skill is used on that repository.

git-analyzer