The Agent Skills Directory

[SAFE]: The skill does not exhibit any malicious behavior such as prompt injection, persistence mechanisms, or unauthorized privilege escalation. It adheres to established development practices for AI agent extensions.
[DATA_EXFILTRATION]: The instructions include a clear security mandate to never include secrets, credentials, or machine-specific private data in the generated skill files. This is reinforced in the core contract and validation rules.
[PROMPT_INJECTION]: While the skill processes external project data (representing an indirect prompt injection surface), it mitigates this risk through defensive prompting, strict formatting requirements (kebab-case names, specific description triggers), and a 12-point quality rubric that includes manual verification before installation.
[COMMAND_EXECUTION]: The documentation references the platform's dynamic context injection syntax (!command) for legitimate use cases. The examples provided (e.g., checking git branches or reading package dependencies) are common development workflows and do not present a security risk in this context.

sigil