The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script roles.py within the <pstack-runtime> directory to manage reviewer roles and subagent delegation. This execution is part of the skill's core orchestration logic and uses controlled environment variables for paths. Evidence is found in Step 3 of SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data in the form of code diffs and file contents to be reviewed by subagents. Ingestion points: Step 1 in SKILL.md gathers file contents and git diff output. Boundary markers: The reviewer-prompt.md template defines a read-only context and explicitly warns reviewers not to edit files, commit, or push. Capability inventory: Subagents are granted file-read and repository search permissions as documented in reviewer-prompt.md. Parent agent uses delegate_task to spawn sub-processes. Sanitization: The skill relies on instructional boundaries and read-only task descriptions rather than programmatic sanitization of the reviewed code.

interrogate