The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted code files for refinement (Ingestion points: target directory/files). It addresses safety by using subagent isolation for each quality pass to prevent context accumulation (Boundary markers present). The skill's capabilities include reading and writing files and executing tests via Bash (Capability inventory). No explicit sanitization is performed on the code content (Sanitization absent), which is expected for code-polishing workflows.
[COMMAND_EXECUTION]: The correctness pass involves running local test suites using the Bash tool (e.g., pytest). This is an essential and expected capability for a tool designed to verify and refine code behavior.

dorodango