The Agent Skills Directory

[SAFE]: The skill design adheres to safety principles by requiring manual confirmation before starting implementations and restricting actions on sensitive branches (main/master) without explicit user consent.
[PROMPT_INJECTION]: The skill processes implementation plans from external files, creating an indirect prompt injection surface. This is effectively mitigated by mandatory 'critical review' instructions (Step 1.1, 1.2) and a requirement to raise concerns to a human partner (Step 1.3) before execution.
[COMMAND_EXECUTION]: While the skill executes steps from a plan, it includes defensive instructions to stop execution immediately upon encountering blockers, dependency issues, or validation failures (Step 2 and 'When to stop' section), preventing uncontrolled command execution.

executing-plans