The Agent Skills Directory

[SAFE]: The skill incorporates robust safety procedures for code modification, including mandatory baseline checks, incremental verification, and a requirement to stop and ask for approval when baseline checks are failing or the scope is broad.
[PROMPT_INJECTION]: The skill processes untrusted input in the form of repository code and documentation, creating a surface for indirect prompt injection.
Ingestion points: Reads project files and structural metadata using Read, Glob, and Grep (SKILL.md).
Boundary markers: Not explicitly defined; the instructions do not mandate specific delimiters or warnings to ignore instructions embedded within the code being refactored.
Capability inventory: The skill utilizes Bash for running diagnostic tools and Write for applying code changes across the filesystem.
Sanitization: No explicit filtering or sanitization of ingested file content is performed prior to processing.
[SAFE]: The skill follows the principle of least privilege regarding environment changes, explicitly instructing the agent not to install new dependencies or tools without prior user consent.

code-refactor