The Agent Skills Directory

[SAFE]: The skill instructions prioritize behavior-preserving transformations and include explicit 'Stop Conditions' that require user intervention for high-risk operations, such as deleting exported symbols or performing broad architectural changes.
[EXTERNAL_DOWNLOADS]: While the skill references various third-party development tools like ast-grep, rope, and jscodeshift, it specifically instructs the agent not to install new tools automatically and to prefer project-native utilities.
[COMMAND_EXECUTION]: The skill uses tools like Bash and Write for legitimate development tasks. The instructions guide the agent to use these for running existing project tests and performing targeted code modifications rather than arbitrary execution.
[PROMPT_INJECTION]: No evidence of prompt injection, role-play bypasses, or instructions to ignore safety guidelines was found in the skill metadata or body.
[DATA_EXFILTRATION]: The skill does not contain patterns for unauthorized data access or external network calls to non-standard domains. It focuses on local repository analysis and modification.

code-refactor