The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/validate_evidence_capsule.py uses subprocess.run() to execute git commands (git show, git rev-parse). These calls are performed with shell=False (default) and use strictly validated, repository-relative paths or Git revisions, following best practices for safe system interaction.
[DATA_EXFILTRATION]: No network operations or external data transmissions were detected. The validate_evidence_capsule.py script is explicitly read-only and uses only the Python standard library, preventing external data flow.
[PROMPT_INJECTION]: The SKILL.md contains strict independence guidelines that instruct the agent to ignore any self-scoring or safety claims produced by the content being audited. It explicitly mandates a fresh session and verification against repository authority, which serves as a defensive measure against malicious input in audited transcripts.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute remote code. All logic is contained within the local Python script and standard Git operations. The independence contract in SKILL.md strictly forbids installing tools or dependencies.

audit-onboarding-proposal