The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies extensively on executing the crit CLI tool to perform various tasks.
Evidence: Commands include crit story --guide, crit story --prep, crit story --story-file, crit, crit comments, and crit comment --reply-to.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it is instructed to process and follow instructions contained in external review comments.
Ingestion points: Instructions specifically tell the agent to 'Read the review output' from crit (Step 6) and 'Address each review comment' (Step 7). These comments originate from the crit tool/service.
Boundary markers: There are no specific delimiters or warnings mentioned to distinguish between legitimate feedback and potentially malicious instructions embedded in the comments.
Capability inventory: The agent has the capability to modify source files (Step 7.3) and execute shell commands (crit CLI).
Sanitization: No sanitization or validation of the comment content is mentioned before the agent acts on the instructions within them.

crit-story