---
name: interrogate
description: "Use for \"interrogate\", \"adversarial review\", \"multi-model review\", \"challenge this\", \"stress test this code\", \"find blind spots\", or \"tear this apart\". Multiple LLM reviewers challenge changes from independent angles."
disable-model-invocation: true
---

# Interrogate

Spawn one reviewer per configured model to adversarially review code changes. Each model gets the same prompt and rubric. The adversarial signal comes from model diversity, not assigned personas.

The deliverable is a synthesized verdict. Do NOT auto-apply changes.

## Step 1, Determine Scope

Identify what to review from context:

- If the user points at specific files or a diff, use that
- If on a feature branch, run `git diff main...HEAD` (or the appropriate base branch) for the full changeset
- If the user's message references recent work, gather the relevant files

Package the diff (or file contents) plus any surrounding context files the reviewers need to understand the code.

## Step 2, State the Intent

Before spawning reviewers, state the intent explicitly. Derive this from:

- The user's message
- Commit messages
- PR description if one exists
- The code itself

Write one clear paragraph. If you're unsure about the intent, ask the user before proceeding.

## Step 3, Spawn Reviewers

Use [session capabilities](../../references/host-capabilities.md), [model selection](../../references/model-roles.md), and the [assignment contract](../../references/orchestration-contract.md). Launch authorized independent read-only reviewers together through the exposed worker operation, one per explicitly configured reviewer selection. Extend or shrink Reviewer A/B labels to the selection count. Give each an immutable assignment, the same input revision and separate result ownership. Inspect actual results before synthesis.

The canonical source used a local model rule and default requests from the Claude and Grok families. The exact historical defaults are preserved as provenance in the [adaptation record](../../docs/parity/adaptations/interrogate.json). These historical requests do not establish available models. Preserve explicit session selections. With no configured selections, request two inherited reviewers and disclose that inherited or identical models do not establish model diversity. For auto or inherit-parent, omit explicit model selection. Never treat aliases as rejected model names.

Select explicit models and effort only from current exposed tool support. If a configured request is rejected, disclose it and use an authorized supported alternative, preferring the same family and reasoning tier where available. If the exact selection is required, report blocked. Record requested and observed models separately; missing observations remain unknown. Do not open a PR or change settings to repair upstream defaults without explicit authorization.

When independent workers are unavailable, apply the same prompt and rubric sequentially with authorization and label reduced capability. Do not call repeated passes independent reviewers or consensus. Missing observed model diversity limits that claim even when worker independence succeeds. Denied review or missing source blocks grounded findings. Do not install tools or configure agents as an incidental step.

Read `references/reviewer-prompt.md` and fill in the template with:
1. The stated intent
2. The diff or file contents
3. The review rubric from `references/rubric.md`
4. The code-quality lens from `references/code-quality-review.md`

The same filled template goes to all reviewers, so every model applies the code-quality lens.

## Step 4, Synthesize

As results come back, build a unified picture:

1. **Parse all findings** from the reviewers
2. **Identify consensus**. Findings raised by 2+ models independently are highest signal.
3. **Identify lone-model findings**. Still worth reading, but weight accordingly.
4. **Deduplicate**. Different models may describe the same issue differently. Merge these and note which models raised it.
5. **Note disagreements**. If one model flags something and another explicitly says the opposite, that's useful context for the verdict.

## Step 5, Lead Judgment

You are the lead reviewer, a pragmatic senior engineer, not a neutral aggregator.

Read `references/lead-judgment.md` for the full framework.

Categorize every finding using these buckets:

- **Act on**. Real issues affecting correctness, security, or maintainability given the actual goals. These would block a real PR.
- **Consider**. Legitimate points, but you're not sure they outweigh the cost of addressing them right now. Worth the user's attention.
- **Noted**. Technically valid but not actionable. Context-dependent, premature optimization, or low-impact given the current stage.
- **Dismissed**. Wrong, nitpicky, or missing context. Brief explanation why.

For each finding, include:
- Which model(s) raised it
- The category (act on / consider / noted / dismissed)
- A one-line rationale for the categorization

## Output Format

Present the verdict in this structure:

### Intent
> [The stated intent paragraph from Step 2]

### Reviewers
Report requested and observed models, worker independence, model diversity and capability limits. Missing observations remain unknown.

- Reviewer [label]: [model name], [N findings] (one bullet per reviewer)

### Act On
[Findings that should be addressed. For each: description, which models raised it, why it matters.]

### Consider
[Findings worth thinking about. For each: description, which models raised it, tradeoff involved.]

### Noted
[Valid but low-priority. Brief list.]

### Dismissed
[Rejected findings with brief rationale.]

### Agreement Map
[Where did models agree, where did they diverge, and what does the pattern of agreement/disagreement tell us?]