The Agent Skills Directory

[Usage and Workflow Constraints]: The skill contains explicit instructions to avoid automatic invocation and restricts its operations to standalone, read-only verification. By requiring an explicit user request and forbidding repository modifications (such as commits or file changes), the skill maintains a controlled execution environment.
[Data Ingestion Surface]: The skill is designed to process external security finding descriptions and reference files. While this involves ingesting untrusted data, the skill mitigates risk by enforcing a strict JSON output contract, which limits the potential impact of instructions embedded within finding reports.
[Controlled Command Execution]: In 'standard' verification mode, the instructions allow for running existing reproducers or regression checks. The skill specifically mandates that these actions must not modify the repository, ensuring that verification remains non-destructive and aligned with the intended security posture.

verify-fix