The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill framework defines a methodology for ingesting untrusted external source code and manifests, then using the findings to generate persistent 'memories' and 'lessons' in the project filesystem. It lacks explicit boundary markers or instructions to sanitize or ignore instructions embedded within the analyzed data, which creates a potential surface where malicious code being audited could influence the agent's logic during internal critique or memory-driven context retrieval.
Ingestion points: The skill analyzes source code, dependency manifests, and architecture diagrams as entry points for findings.
Boundary markers: Absent; reference files do not specify delimiters or 'ignore instructions' warnings when processing external codebase data.
Capability inventory: The skill performs file writes to the .github directory and manages automated tool retry and refinement flows.
Sanitization: No explicit logic is provided to filter or escape findings derived from untrusted content before they are written to the self-learning memory system.

audit-integrity