The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources to reconcile project status, creating a vulnerability surface for indirect prompt injection. 1. Ingestion points: The skill reads data from docs/*.md, gh issue list, ROADMAP.md, CHANGELOG.md, and README.md as specified in reference/source-triangulation.md. 2. Boundary markers: The instructions lack explicit boundary markers or instructions to disregard embedded commands in the ingested artifacts. 3. Capability inventory: The agent has the ability to search code files, read documentation, and interact with the gh CLI for metadata retrieval. 4. Sanitization: No sanitization or validation protocols are described for the external content before it is incorporated into the agent's context or reasoning process.

pdm