The Agent Skills Directory

[SAFE]: The skill follows a disciplined methodology for code modification, utilizing a local harness script (orchestrate.mjs) to create rollback points (git stashes) before any edits are made. This ensures that any change is reversible.
[COMMAND_EXECUTION]: The skill executes local commands through a harness script to manage the refactoring lifecycle. These commands include checkpoint to save state, advance to commit verified changes, and fail to trigger self-healing rollbacks. All operations are confined to the local repository and the skill's own directory.
[PROMPT_INJECTION]: The skill is designed to process and modify user-provided source code, which constitutes a surface for indirect prompt injection.
Ingestion points: Source code files within the user's repository (SKILL.md, references/method.md).
Boundary markers: The instructions do not specify explicit delimiters or warnings to ignore instructions embedded in the code being refactored.
Capability inventory: The skill can modify file contents and execute shell commands via the orchestrate.mjs script.
Sanitization: No explicit sanitization of input code is performed. However, the mandatory 'identical pass-set' verification step acts as a robust control, as any malicious behavior injected via code comments or logic would likely alter the program's behavior and trigger an automatic rollback.

refactor-transform