The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from the inbox/ directory, which presents a surface for indirect prompt injection where malicious instructions in the source text could influence the resulting knowledge or skill generation.
Ingestion points: Reads contents of files in inbox/ that have a pending status.
Boundary markers: Does not implement explicit delimiters or instructions to ignore embedded prompts for the agent when processing external content.
Capability inventory: The skill can create, update, and merge files within the knowledge base and registry, and can initiate cross-calls to other skills like pk-capture and pk-track.
Sanitization: Relies on mandatory user review and explicit approval (Step 4 and Step 5.5) before any proposed changes or skill promotions are executed.
[COMMAND_EXECUTION]: The skill invokes local shell commands to maintain the integrity and structure of the knowledge base.
Uses grep -ril to perform case-insensitive recursive searches for duplicate content across local directories.
Uses grep -rl to identify and rewrite inbound links (backlinks) when merging or deprecating knowledge pages.
These operations are limited to the local filesystem within the defined .cockpit/ directory structure.

pk-distill