---
name: audit-feedback-loop
description: Coordinate iterative evidence and reliability reviews between BrainPilot's Principal Investigator and Auditor. Use when PI needs to audit its own draft, an Expert result, or a multi-agent synthesis; when Auditor receives such a review task; or when a previous audit requires correction and incremental re-review.
---

# Audit Feedback Loop

Use the durable task system for every handoff. Do not ask the user to relay audit findings.

## Select the role procedure

- If you are `principal`, read [references/pi-orchestration.md](references/pi-orchestration.md) and follow it. Load the request template before dispatching an audit. Load the revision-loop reference after receiving a `revise` or `block` verdict.
- If you are `auditor`, read [references/auditor-review.md](references/auditor-review.md) and follow it. Load only the specialist audit skills relevant to the assigned risks, then load the response template before completing the task.
- If you are any other agent, provide an evidence-addressable result to PI; do not start or adjudicate the audit loop yourself.

## Load only the needed templates

- Audit request: [references/audit-request-template.md](references/audit-request-template.md)
- Audit response: [references/audit-response-template.md](references/audit-response-template.md)
- Correction and incremental re-review: [references/revision-loop.md](references/revision-loop.md)

Keep audit requests in `dispatch_task`, full findings in the versioned report,
and only the compact report pointer in `complete_task`. Do not use a separate
audit-report submission tool.

## Specialist audit skills

- [Data integrity](../audit-data-integrity/SKILL.md)
- [Model validation](../audit-model-validation/SKILL.md)
- [Code and artifact](../audit-code-artifact/SKILL.md)
- [Evidence](../audit-evidence/SKILL.md)