The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/scan.py script uses subprocess.run to orchestrate repository checks. It executes git commands to list files and retrieve content blobs, and it triggers existing project tests using uv run pytest and bash. These operations are strictly limited to local repository management and testing.
[DATA_EXPOSURE]: The skill is designed to scan for and identify sensitive information such as AWS keys, GitHub tokens, and PII. It accesses sensitive paths like .env and credentials/ to perform its audit. To mitigate risk, the scanner implements a redaction mechanism (_redact_snippet) that masks detected secret values before displaying them in reports.
[INDIRECT_PROMPT_INJECTION]: The scanner ingests untrusted data from repository files and presents it to the agent.
Ingestion points: Reads arbitrary file contents and git diffs in scripts/scan.py.
Boundary markers: Results are structured using text, markdown, or JSON formatting.
Capability inventory: Subprocess execution (git, pytest, bash) and file writing (.gitignore remediation).
Sanitization: Secret patterns are redacted in the output snippets. The SKILL.md instructions specifically direct the agent to wait for human confirmation for critical findings.

compliance-scan