The Agent Skills Directory

[SAFE]: No malicious patterns or security risks were identified. The skill follows best practices for repository documentation management, including explicit instructions to keep sensitive data out of documentation files.
[PROMPT_INJECTION]: The skill processes untrusted repository data (docs, manifests, file lists) which presents a surface for indirect prompt injection. This is addressed through instructions that prioritize human-readable summaries and require explicit user approval for any modifications to the root instructions file.
Ingestion points: Reads root AGENTS.md, files within docs/, package manifests, and repository file structure via rg --files.
Boundary markers: Snippet integration uses  markers to maintain clear separation of methodology content.
Capability inventory: File creation and updates for documentation in docs/ and instruction updates in AGENTS.md.
Sanitization: The workflow mandates providing semantic summaries and diffs for user review before any persistent changes are made, particularly to the agent's core instruction file.

project-memory