The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes code changes and diffs provided by the user or read from the filesystem and interpolates them directly into prompts for delegated reviewer tasks via the template in references/reviewer-prompt.md. Maliciously crafted comments or code within these files could contain instructions designed to deceive the reviewer agents or bypass their safety guidelines.
Ingestion points: Code diffs and file contents are gathered in SKILL.md Step 1.
Boundary markers: The references/reviewer-prompt.md template lacks boundary markers or explicit instructions for reviewers to ignore embedded prompts within the {DIFF_OR_FILES} block.
Capability inventory: The skill uses delegate_task to spawn multiple sub-agents and git diff to read files.
Sanitization: There is no evidence of sanitization or escaping of the ingested code content before it is interpolated into the prompt.
[COMMAND_EXECUTION]: The skill directs the agent to execute shell commands, specifically git diff, to gather code context. There is a potential risk if the branch names or file paths are derived from unvalidated user input.
Evidence: SKILL.md instructs the agent to run git diff main...HEAD or determine an "appropriate base branch" based on the user's message, and to execute kickoff-setup if certain configurations are missing.

interrogate