interrogate

Interrogate

Spawn four reviewers to adversarially review code changes. Each gets the same prompt and rubric. The adversarial signal comes from model diversity, not assigned personas. Models differ in blind spots, priors, and reasoning patterns. Agreement across models is high-confidence signal; lone-model findings are worth reading but lower confidence.

Adapted for Claude Code. The Agent tool exposes three Claude tiers (opus, sonnet, haiku) from a single provider, not four cross-provider models. The reviewers below vary by tier and by the prompt, so true cross-provider diversity is reduced relative to upstream pstack. Weight cross-model agreement accordingly and lean harder on the lead judgment in Step 5. If you have other providers wired up as subagent types or MCP-backed reviewers, add them for real diversity.

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. What is this code trying to accomplish? Derive this from:

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

Write one clear paragraph. Reviewers challenge whether the work achieves the intent well, not whether the intent itself is correct. If you're unsure about the intent, ask the user before proceeding.

Step 3, Spawn Reviewers

Launch all four in a single message using the Agent tool, each with a different model.

Subagent	Model
Reviewer A	opus
Reviewer B	sonnet
Reviewer C	haiku
Reviewer D	sonnet

For each reviewer:

• subagent_type: general-purpose
• model: the model from the table
• read-only posture: the prompt forbids file writes (the Agent tool has no readonly flag)

If a model slug in the table is rejected as unresolvable when you try to spawn the subagent, check the valid slugs in the Agent tool's error message, pick the closest equivalent (prefer the highest-reasoning tier of the same family), spawn with the valid slug, and open a separate PR to update this table. Do not block the review on the slug issue.

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 four reviewers, so every model applies the code-quality lens.

Each reviewer produces structured findings as described in the prompt template.

Step 4, Synthesize

As results come back, build a unified picture:

1. Parse all findings from the four 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. Reviewers only see a slice of the codebase. You have the full context (the goal, the constraints, the timeline, which tradeoffs were already considered). Use that context aggressively.

Categorize every finding into one of four 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

• Model A: [model name], [N findings]
• Model B: [model name], [N findings]
• Model C: [model name], [N findings]
• Model D: [model name], [N findings]

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. This shows the user what was filtered out and why, so they can override your judgment if they disagree.]

Agreement Map

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