The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/repo_maintainer.py executes git commands via subprocess.run to perform repository analysis.
Evidence: The run_git and disabled_filter_config functions call the Git CLI to inspect repository state and diffs.
Mitigation: The script includes extensive security hardening, such as disabling potentially malicious Git filters (clean, process, required), sanitizing environment variables (stripping GIT_ prefixed variables), and using list-based arguments for subprocess to prevent shell injection.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from repository changes, creating a surface for indirect prompt injection.
Ingestion points: scripts/repo_maintainer.py reads file contents and Git diff outputs directly from the repository being maintained.
Boundary markers: The instructions in SKILL.md advise treating analyzer results as conservative signals and emphasize independent verification of results by the agent.
Capability inventory: The skill can read any file in the repository, execute Git commands, and write manifest files to disk.
Sanitization: The analyzer includes markdown_escape and markdown_code functions to sanitize data rendered in reports, preventing the injection of formatting or bidirectional control characters into the agent's context.
[DYNAMIC_EXECUTION]: The test suite utilizes dynamic module loading for local script verification.
Evidence: tests/test_repo_maintainer.py uses importlib.util.spec_from_file_location and module_from_spec to load the repo_maintainer.py script for unit testing.
Context: This is a standard Python testing practice used to verify local code and does not involve the execution of untrusted remote content.

repo-maintainer