The Agent Skills Directory

[SAFE]: The skill is entirely instructional and operates locally on Markdown files within a defined vault structure. It does not perform network operations, execute shell commands, or include external dependencies.
[SAFE]: The instructions include robust safety practices, such as requiring a review_required: true flag on all generated pages, preserving human-owned sections, and prohibiting the deletion of source material.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources (PDFs, articles, web captures) which could potentially contain malicious instructions.
Ingestion points: Files located in the raw/ and inbox/ directories.
Boundary markers: The skill employs a draft lifecycle and explicitly marks AI-generated content with metadata, though it does not define specific prompt delimiters for raw content.
Capability inventory: Limited to read/write access for Markdown files in specific subdirectories (wiki/sources, wiki/entities, etc.). It lacks dangerous capabilities like network exfiltration or code execution.
Sanitization: The procedure relies on manual review and claim classification rather than automated input sanitization.

wiki-ingest