The Agent Skills Directory

[SAFE]: The skill identifies and processes local files in the raw/ directory, ensuring all operations remain within the project's scope and do not involve unauthorized data access or exfiltration.
[SAFE]: A mandatory human-in-the-loop safeguard is implemented, requiring the agent to present findings and wait for user confirmation before creating or updating any wiki pages. This prevents the agent from making autonomous changes based on external inputs.
[SAFE]: The skill references the qmd tool for search functionality but explicitly instructs the user to handle installation and index updates, avoiding the autonomous execution of potentially dangerous setup scripts.
[SAFE]: The skill manages an indirect prompt injection surface when ingesting external data.
Ingestion points: Local files in the raw/ directory and content processed via pdf-docling or video-whisper MCP tools.
Boundary markers: Explicit instructions to wait for user confirmation before proceeding with wiki updates.
Capability inventory: Reads local files and creates or updates markdown files in defined wiki subdirectories.
Sanitization: Security relies on the mandatory manual verification step by the user.

wiki-ingest