The Agent Skills Directory

[SAFE]: The skill includes a security policy in agents/openai.yaml that disables allow_implicit_invocation. This prevents the skill from being triggered automatically by an agent, requiring explicit user intent and mitigating risks of recursive execution or unauthorized codebase mutations.
[SAFE]: Instructional templates in SKILL.md employ clear boundary markers, such as specific headers and markdown code blocks, to isolate user-provided data (e.g., error logs, symptoms) from the agent's instructions. This structure helps prevent the agent from misinterpreting raw error output as new commands.
[SAFE]: The skill provides explicit guidance to evaluate code with rigor and not assume its correctness, particularly when dealing with generated code, which promotes safer development practices.

claude-debug