Security audit
Security checks for vulnerabilities and agentic risk
This is a read-only benchmark-audit skill with a disclosed but notable GitHub PR lookup and parallel subagent workflow.
Installers should understand that this skill is read-only, but not strictly offline: when GitHub CLI is available it may inspect open PR metadata and diffs for the target repository, potentially using the user's existing GitHub session. Run it only on repositories where that access is acceptable, and prefer the sequential fallback if parallel subagents would be too costly or heavy for the environment.
The skill is framed as a read-only local repository audit, but it instructs the agent to query open PRs with gh pr list and inspect remote diffs. That expands the trust boundary from the local checkout to network-accessible repository state, which can leak metadata, surprise users, and violate least-privilege or offline-expectation assumptions even if no files are modified.
The contradiction is not about file modification, but about stated audit scope. The text frames the skill as a read-only benchmark-repo audit, while later mandatory steps require reviewing GitHub PR state and diffs, meaning the actual process includes remote change-review activity beyond the immediate repo contents.
This markdown file instructs the user to run multiple subagents in parallel, which can affect system load, cost, or produce concurrent actions, but it does not include any caution about those operational effects. Under the markdown variant of SQP-2, descriptions should warn when behaviour could affect system integrity or user resources.
No suspicious patterns detected.
