The Agent Skills Directory

[SAFE]: The skill implements a rigorous peer-review methodology designed to detect errors or omissions in existing analyses. It emphasizes read-only access to production code and requires that any tests or probes be conducted in an isolated environment.
[INDIRECT_PROMPT_INJECTION]: While the skill processes external assessments that could contain malicious instructions, the provided 'Doubt discipline' explicitly instructs the agent to separate persuasive narratives from technical grounding and to verify findings against constraints. This structured approach serves as a defensive boundary against injection attempts in processed data.
[COMMAND_EXECUTION]: The instructions permit the execution of 'isolated probes' or tests to verify findings. The skill limits the impact of these operations by mandating they run in a scratch/isolated state and ensuring product code remains read-only, following least-privilege principles.

challenge