The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of various local shell commands to manage the refactoring lifecycle. This includes using git for version control, grep for code analysis, and a specific Python script validate-yaml.py to verify the integrity of generated checklists. It also executes project-specific tests, linters, and type checkers during the verification phase.- [PROMPT_INJECTION]: The skill is designed to ingest and modify untrusted data in the form of user source code. While this is necessary for its primary function, it creates a potential attack surface for indirect prompt injection where malicious instructions embedded in code could influence the agent's behavior.
Ingestion points: The agent reads source files from the repository during the scan and apply phases to identify optimization points and perform modifications.
Boundary markers: The workflow utilizes structured Markdown files and YAML frontmatter (e.g., scan.md, refactor-design.md) to distinguish between operational instructions and the code being processed.
Capability inventory: The skill possesses significant capabilities, including file write access across the repository, execution of shell commands (git, grep, python), and retrieval of file contents.
Sanitization: There is no automated sanitization of the processed code. Instead, the skill relies on a rigid multi-stage human approval process (requiring user sign-off for the scan results, the design plan, and every individual execution step) to mitigate the risk of unintended or malicious changes.

cs-refactor