The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run across several scripts (submit_contribution.py, star_repo.py) to interface with git and the GitHub CLI (gh). These calls are used to automate forking, committing, and opening pull requests. Security is maintained by passing commands as argument lists rather than raw shell strings, which prevents shell injection vulnerabilities.
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file utilizes the platform's dynamic context syntax (!command) to trigger bundled Python scripts when the skill is accessed. These scripts perform local tasks such as identifying modified files and checking safety status. The commands are scoped to the skill's own utility scripts and are used to populate report artifacts for the user's review.
[CREDENTIALS_UNSAFE]: A hardcoded dummy GitHub token is present in tests/test_contribution_safety.sh. This string (ghp_abcdef...) is a non-functional placeholder used for regression testing the safety scanner's ability to detect secrets and does not constitute a credential leak.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of local skill modifications. It employs a robust mitigation strategy including:
Ingestion points: Local skill files read by find_local_changes.py.
Boundary markers: Explicit instructions for the agent to never send data without line-by-line user confirmation.
Capability inventory: File reading and GitHub submission capabilities via git and gh.
Sanitization: A specialized safety scanner (check_contribution_safety.py) that uses extensive regex patterns to detect and block Personal Health Information (PHI) and active credentials, alongside hash verification in submit_contribution.py to ensure only scanned content is submitted.

contribute