In this article
Finding Deep Verifier
| Field | Value |
|---|---|
| Kind | agent |
| Source | .github/agents/security/subagents/finding-deep-verifier.agent.md |
| Invocation | Delegated subagent, dispatched by a parent agent (not selected directly) |
| Interactive | No |
What it does
Deep adversarial verification of FAIL and PARTIAL findings for a single security skill
When to use it
Reviewer orchestrators dispatch this worker to verify all FAIL and PARTIAL findings for one skill or supported RAI framework in audit or diff mode. It searches for confirming and contradicting evidence rather than accepting an assessor's claims unchanged. Plan mode skips this verification stage.
Example usage
The parent supplies the skill, profile, non-empty candidate findings, and diff context for an authorization change. The worker returns a verification verdict and supporting evidence for each finding. A RAI dispatch explicitly supplies Domain=rai and the framework contract; unsupported domains are rejected. The parent decides how to consolidate the results, and the worker does not implement suggested fixes.