Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed, read-first skill audit tool that only recommends skill cleanup actions unless the user later chooses changes.
Install this if you want help auditing and cleaning up local skills. Expect it to inspect the current thread and visible skill directories, and review any proposed Fix, Merge, or Delete action before authorizing changes.
The manifest description says to use the skill whenever the user wants to "audit, optimize, clean up, consolidate, or improve" skill-related quality. These are broad, everyday terms and the file does not provide explicit trigger phrases, exclusions, or negative examples to limit when the skill should activate.
The default prompt describes activation as auditing the current conversation and local skills, but it does not define specific trigger phrases, scope limits, or exclusion conditions. Terms like "audit" and "generate an actionable cleanup plan" are broad enough that the skill could be invoked for general review requests rather than a narrowly defined context.
The prompt asks to audit 'current workspace' skills and infer missed triggers from conversation evidence, but it does not define clear activation boundaries or user-consent constraints. In a skill-routing system, broad language like this can cause over-triggering, unnecessary inspection of local skills, and expansion of scope beyond what the user explicitly intended.
The request for a 'governance-style audit report' is ambiguous about whether it should analyze only the current thread, all installed skills, or broader local state. That ambiguity is risky because it can lead the skill to activate in contexts where the user did not clearly authorize broad inspection, increasing privacy and overreach concerns.
The action-queue prompt uses very common operational language like 'fix, merge, delete, or keep and skip' without strong trigger specificity, which increases the chance that ordinary user requests could invoke this skill unintentionally. While the output is constrained, accidental activation could still steer users toward sensitive skill-management actions based on incomplete or incidental context.
No suspicious patterns detected.
