The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/verification_context.py executes local git commands via subprocess.run. The implementation follows security best practices by using argument lists rather than shell strings, effectively preventing command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted implementation data including git diffs and task summaries. Ingestion points include the verification_context, task_summary, and claimed_validation variables in references/prompt-template.md. The skill uses markdown code blocks as boundary markers but does not perform active sanitization. However, the risk is mitigated by the agent's instructions to challenge completion claims and maintain a critical review posture. The skill's capability inventory is limited to local repository inspection and read-only review.

verification-gate