The Agent Skills Directory

[SAFE]: The skill serves as a safety and compliance auditor for other skills. It does not perform autonomous actions and requires explicit user approval for any proposed changes to the filesystem.
[COMMAND_EXECUTION]: The skill mentions external tools such as 'git' and 'npm', but only within the context of analyzing project history or recommending commands for the user to run manually. It specifically instructs the agent not to execute build commands automatically.
[INDIRECT_PROMPT_INJECTION]: The skill processes 'external skills' provided as input. While this constitutes an ingestion surface for untrusted data, the skill is designed to mitigate risks by transforming these inputs into a highly structured format that wraps instructions in safety guardrails and approval gates.
[PRIVILEGE_ESCALATION]: No privilege escalation patterns were detected. The skill operates within the standard scope of skill management and emphasizes 'least privilege' by using phase-gated file limits for edits.
[DATA_EXPOSURE]: The skill reads existing skill definitions from the '.agents/skills/' directory. This is an intended administrative function and no evidence of exfiltration to external domains was found.

duck-adapt