The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes project-specific scripts for formatting, linting, and testing (e.g., in Step 2). This is standard behavior for code-refinement tools but involves executing commands defined within the target project's environment.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests and processes untrusted project code.
Ingestion points: Project code is ingested via git diff --cached in Steps 3, 4, and 7 of SKILL.md.
Boundary markers: No explicit boundary markers or 'ignore' instructions for embedded data are defined in this orchestration skill.
Capability inventory: The skill can execute shell commands (formatter/linter/tests), invoke other skills (Step 1-6), and delegate tasks to subagents (Step 7).
Sanitization: There is no evidence of sanitization or filtering of the ingested code before it is passed to other skills or subagents.
[REMOTE_CODE_EXECUTION]: Step 7 delegates smoke test execution to a subagent using the Agent tool (model: "opus"), which involves the subagent potentially executing code to verify changes.

polish-code