The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute several local development tools and scripts to verify code integrity and formatting.
Evidence: References to running validate-yaml.py (SKILL.md), automated tests (SKILL.md, methods.md), type checkers, and linters (SKILL.md).
Context: These are standard development practices within the refactoring workflow and are used for verification rather than arbitrary execution.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted local source code to generate refactoring suggestions.
Ingestion points: The scan phase reads project files and directories (SKILL.md).
Boundary markers: The workflow uses structured document templates and status fields (e.g., status: draft) to delimit content.
Capability inventory: The agent has the ability to read and write files, and execute local shell commands for testing and linting.
Sanitization: No explicit technical sanitization is mentioned; however, the workflow mandates human review (HITL) at every stage (scan, design, and apply), which acts as the primary defense against malicious instructions embedded in the code being analyzed.
[SAFE]: The skill incorporates a robust 'Refusal Routing' system that proactively identifies and rejects requests that fall outside the safety scope, such as those missing test coverage or involving behavioral changes.

cs-refactor