The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands to perform verification gates and manage version control during the refactoring process.
Evidence: Automated detection and execution of project-specific test runners, linters, and type checkers (e.g., npm test, pytest, npx tsc, cargo test).
Evidence: Use of git for staging changes, committing with specific messages, and reverting failed chunks via git checkout or git restore.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect injection as it processes external, untrusted assessment documents to generate action plans.
Ingestion points: Phase 1 involves reading user-provided assessment documents, such as markdown files or output from tools like seam-ripper or simplicity audits (SKILL.md).
Boundary markers: While the skill uses a structured refactor-plan.md as an intermediary to separate input from execution, it does not explicitly define delimiter-based boundary markers for the ingested assessment text.
Capability inventory: The skill has broad capabilities including arbitrary file system writes to the codebase and the execution of shell commands via the verification suite.
Sanitization: The skill relies on the agent's extraction logic to transform recommendations into a plan, which serves as an implicit sanitization layer, but no explicit data escaping is implemented.

architectural-refactor