The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection surface identified.
Ingestion points: External source files (transcripts, PDFs, markdown) located in the raw/ directory.
Boundary markers: No specific delimiters or boundary markers are used for source content ingestion.
Capability inventory: Writing and updating files in the wiki/ directory, updating local index files, and executing search commands.
Sanitization: No explicit sanitization or filtering of source content is described.
Mitigation: The skill implements a mandatory human-in-the-loop checkpoint where extracted points and matched pages are presented to the user for confirmation before any modifications occur, neutralizing the risk of unintended behavior from malicious source data.
[COMMAND_EXECUTION]: The skill utilizes local CLI tools (grep, qmd) for content discovery. It follows security best practices by explicitly instructing the agent not to run indexing or embedding commands automatically, requiring manual user execution for state-changing operations.

wiki-ingest