The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a helper script scripts/change_map.py that utilizes subprocess.run to execute git and ripgrep (or grep) for analyzing code changes and finding symbol references. The script passes a base reference derived from external input directly to git diff.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external, potentially untrusted sources including pull request descriptions, linked issues, existing comments, and repository-specific specification files.
Ingestion points: Data enters the agent's context through tools that fetch pull request details (gh pr view), diffs, and by reading local project files like .scratch/<branch>/spec.md and plan.md as described in SKILL.md.
Boundary markers: The skill contains explicit instructions to "Verify every finding in the code" and "Confirm it or drop it," which act as defensive instructions against malicious or incorrect data in the review sources.
Capability inventory: The agent can execute shell commands via the provided analysis script and is specifically instructed to modify files and commit changes to the repository based on its findings.
Sanitization: The workflow does not specify automated sanitization of the external text processed during the review, relying on the agent's analytical steps to verify claims against the actual source code.

deep-review