The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a Python script scripts/repo_maintainer.py that executes local Git commands via subprocess.run to analyze repository history and working tree states.
Evidence: The script contains numerous calls to subprocess.run with the git command.
Security Measures: The script explicitly disables potentially dangerous Git filters (clean, smudge, and process filters) using disabled_filter_config to prevent execution of arbitrary code defined in a repository's local configuration. It also uses argument separators (--) and environment isolation to mitigate command injection risks.
[DYNAMIC_EXECUTION]: The unit test suite tests/test_repo_maintainer.py employs dynamic loading to import the script under test.
Evidence: Use of importlib.util to load repo_maintainer.py from a relative path.
Context: This is a standard practice for testing standalone Python scripts and does not involve executing untrusted code from external sources.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of repository files, commit messages, and Git diffs, which could contain malicious instructions designed to influence the agent's behavior.
Attack Surface: The analyze command reads file contents and Git metadata into the agent's context.
Mitigations: The script implements markdown_escape to sanitize data before rendering reports, preventing Markdown-based injection. Additionally, the skill's instructions (SKILL.md and references/verification.md) emphasize human-in-the-loop verification tiers and manual reconciliation of evidence.

repo-maintainer