interrogate skill
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.
- GitHub stars
- 16
- Files
- 5
- Last commit
- Nov 4, 2026
- License
- MIT
- Allowed tools
- —
What does the interrogate skill do?
interrogate uses 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.
How to install the interrogate skill
Each command copies the skill folder into the client's skills directory. Restart the client or start a new session to load it.Install the interrogate skill in Plugin
/plugin marketplace add arjitj2/open-pstack /plugin install pstack@open-pstack
Install the interrogate skill in Claude Code
~/.claude/skills/interrogate/SKILL.md
$ npx degit arjitj2/open-pstack/plugins/pstack/skills/interrogate ~/.claude/skills/interrogateInstall the interrogate skill in Claude Code (project)
.claude/skills/interrogate/SKILL.md
$ npx degit arjitj2/open-pstack/plugins/pstack/skills/interrogate .claude/skills/interrogateInstall the interrogate skill in Codex
~/.agents/skills/interrogate/SKILL.md
$ npx degit arjitj2/open-pstack/plugins/pstack/skills/interrogate ~/.agents/skills/interrogateInstall the interrogate skill in Gemini CLI
~/.gemini/skills/interrogate/SKILL.md
$ npx degit arjitj2/open-pstack/plugins/pstack/skills/interrogate ~/.gemini/skills/interrogateInstall the interrogate skill in Claude.ai
npx degit arjitj2/open-pstack/plugins/pstack/skills/interrogate interrogate && zip -r interrogate.zip interrogate # claude.ai → Customize → Skills → + → Create skill → Upload a skill → interrogate.zip
SKILL.md
View on GitHubInterrogate
Spawn one reviewer per configured model to adversarially review code changes. The adversarial signal comes from model diversity, not assigned personas. The deliverable is a synthesized verdict. Do NOT auto-apply changes.
Dispatch contract. Read provider-dispatch.md before launching reviewers. Configured entries are provider-qualified descriptors; the parent starts native and external read-only lanes directly. On Codex, resolve remaining Claude tool names via codex-tools.md.
Step 1, 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 with any surrounding context files the reviewers need to understand the code.
Step 2, Intent
Before spawning reviewers, state the intent in one clear paragraph from the user's message, commit messages, the PR description if one exists, and the code. If you're unsure about the intent, ask the user before proceeding.
Step 3, Spawn Reviewers
Start all reviewers in one fan-out phase. Use interrogate reviewers from the current harness's pstack model sheet when present, one reviewer per entry, extending or shrinking the Reviewer A/B/C labels below to the configured entry count. If the sheet or that line is missing, use the table defaults. Native reviewers use the parent subagent primitive. External reviewers use the launcher directly and must return a complete, model-verified receipt.
| Subagent | Default model |
|---|---|
| Reviewer A | codex:gpt-5.6-sol@max |
| Reviewer B | grok:grok-4.7@xhigh |
| Reviewer C | claude:opus@max |
For each reviewer, route the configured descriptor with read-only access and a unique output/receipt path. If the descriptor is inherit-parent or auto, use the parent subagent primitive without a model override. Each reviewer seat runs its saved attempt chain from the model sheet: a terminal outcome the sheet's # fallback policy authorizes advances to the next saved attempt per provider-dispatch's fallback state machine (usage-exhausted only when no policy is saved); every other failure records a dropout and continues with the completed reviewers. Label each finding by the provider and model that actually ran, and say when fallbacks collapse provider diversity — two reviewers on one provider are not cross-provider agreement. Never pick the closest model or fall back outside the saved chain.
Read references/reviewer-prompt.md and fill in the same template for every reviewer with the stated intent, the diff or file contents, the rubric from references/rubric.md, and the code-quality lens from references/code-quality-review.md.
Step 4, Synthesize
As results come back, build a unified picture. Retain native terminal results and external task/session handles and verified receipts. Wait for every configured seat to reach a verified terminal result or recorded dropout before the final lead verdict; process startup, progress, and self-reported completion do not count. Parse every reviewer's full findings. Merge findings that describe the same issue differently, and note which models raised each one. Findings two or more models raised independently are the highest signal. Read a lone model's finding, but weight it accordingly. Note disagreements. If one model flags something and another explicitly says the opposite, that is 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. Rerun each supplied repro command inside a disposable copy of the checkout, never in the working tree, and never against live services, credentials, or the network. Rank a rerun-confirmed finding first in its bucket and label it run-proven. Keep a source- or policy-cited finding that was not run, labeled not run. Put every finding in one of four buckets, with the model(s) that raised it and a one-line rationale.
- Act on. Real issues affecting correctness, security, or maintainability given the actual goals. They would block a real PR.
- Consider. Legitimate, but you're not sure they outweigh the cost of addressing them now. Worth the user's attention.
- Noted. Technically valid but not actionable. Context-dependent, premature optimization, or low-impact at this stage.
- Dismissed. Wrong, nitpicky, or missing context, with a brief reason.
Output Format
Intent
[The stated intent paragraph from Step 2]
Reviewers
- Reviewer [label]: [model name], [N findings] (one bullet per reviewer)
Act On
[Each: description, which models raised it, why it matters.]
Consider
[Each: description, which models raised it, the tradeoff.]
Noted
[Brief list.]
Dismissed
[Each with a brief rationale.]
Agreement Map
[Where models agreed and diverged, and what that pattern tells us.]
Files in this skill
5interrogate skill questions
How do I install the interrogate skill in Claude Code?
Run npx degit arjitj2/open-pstack/plugins/pstack/skills/interrogate ~/.claude/skills/interrogate. Claude Code loads skills from ~/.claude/skills (all projects) or .claude/skills (one project) and uses a skill when its description matches the task.
Which clients can use the interrogate skill?
Any client that reads the Agent Skills format: Claude Code, Codex, Gemini CLI and claude.ai (upload as a ZIP). Scripts in the skill run only where the client allows code execution.