The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is designed to identify and persist 'learnings' and 'patterns' from a session into repository files. If the agent processes untrusted external data during a session, there is a risk that malicious instructions contained within that data could be recognized as a 'learning' and written into the repository's instructions or skills.
[Ingestion Point]: The skill ingests content from the current session's discoveries and major learnings.
[Capability Inventory]: The agent uses file-writing capabilities to modify .github/skills/, .agents/skills/, and .github/instructions/ folders within the repository.
[Boundary Markers & Sanitization]: While the skill includes a manual 'Quality Checks' phase (Step 4) to evaluate if a learning is general and specific, it lacks programmatic sanitization to filter or escape instructions before they are interpolated into the repository files.

update-skills