The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project code which is an untrusted source, creating a surface for indirect prompt injection during the review and refinement phases. Ingestion points: Code diffs are retrieved via git diff --cached (Step 3) and processed through multiple auxiliary skills (/review-code, /evaluate-findings). Boundary markers: The instructions do not specify explicit delimiters or isolation markers when providing code snippets to the review and evaluation skills. Capability inventory: The skill possesses significant capabilities, including file modification (/apply-findings), complex git command execution (Step 5), and the ability to spawn subagents for test execution (Step 6). Sanitization: No sanitization of the codebase content is mentioned before it is processed by the agent or subagents.
[COMMAND_EXECUTION]: The skill utilizes a series of git commands (e.g., git diff, git apply --cached, git hash-object, git checkout) to manage state, apply patches, and verify mutations. These are used for project maintenance and are consistent with the skill's primary purpose.
[DYNAMIC_EXECUTION]: Step 6 involves spawning a subagent (model: "opus") to execute a smoke test plan. While this involves running project code, the instructions include safety protocols such as pre-execution state capture, post-execution verification, and explicit write-access restrictions for the subagent.

polish-code