The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill facilitates the ingestion of untrusted data from PR comments into a persistent library that influences future agent actions.
Ingestion points: PR Review comments and files within the patterns/ directory (as referenced in SKILL.md).
Boundary markers: Absent. There are no defined delimiters or 'ignore' instructions to prevent the agent from obeying malicious commands embedded within PR comments.
Capability inventory: The fix-coordinator agent performs code modifications based on these patterns, and the knowledge-writer agent has file-write capabilities to persist patterns to the filesystem.
Sanitization: Absent. There is no evidence of filtering, escaping, or validating the content of PR comments before they are incorporated into the pattern library.
Risk: A malicious PR comment could contain hidden instructions that, once 'learned' as a pattern, could lead to automated backdoor injection or data exfiltration in subsequent automated repair tasks.

knowledge-patterns