The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to run various test frameworks (e.g., Jest, Pytest, Go test) and an orchestration script (orchestrate.mjs) located within its own package. These executions are scoped to the primary purpose of verifying code behavior and maintaining state during a refactor chain.
[PROMPT_INJECTION]: The instructions do not contain any attempts to bypass safety filters or override agent behavior. The phrases designated as critical, such as 'NEVER refactor on a red repo', are functional constraints intended to enforce development best practices rather than security bypasses.
[DATA_EXFILTRATION]: No evidence of network exfiltration or unauthorized access to sensitive file paths (such as credentials or SSH keys) was found. File system interactions are limited to reading project metadata and running tests in the target directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes external repository files and test outputs, which constitutes an ingestion surface. However, the logic is constrained to identifying test success/failure and characterizing output, with no high-privilege capabilities exposed to the parsed data.

refactor-safety-net