skills.rest
adjohn
adjohnCOMMUNITY
Explore all skills by this author →
GitHub

interrogate

Spawns multiple LLM reviewers to adversarially review code changes and synthesize a verdict.

3|2|Updated Aug 28, 2026
Search Tags:#code quality#diff analysis#code review#pull request#llm#multi-model#adversarial review
One-click install
npx skills add https://github.com/adjohn/pstack --skill interrogate-adjohn
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: interrogate
Source: https://github.com/adjohn/pstack/tree/main/skills/interrogate
Command: npx skills add https://github.com/adjohn/pstack --skill interrogate-adjohn

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Single-reviewer code reviews miss blind spots and produce noisy or one-sided feedback. This Skill runs the same code change through multiple independent LLM reviewers, then applies lead-reviewer judgment to filter findings into an actionable verdict. ## Core Features & Use Cases - Multi-Model Adversarial Review: Spawns one reviewer per configured model tier (fable, sonnet, haiku, opus) so agreement across models signals high-confidence issues. - Structured Rubric and Quality Lens: Every reviewer applies a shared rubric covering correctness, root causes, structural integrity, verification, complexity, and security, plus a code-quality lens targeting structural simplification. - Lead Judgment Synthesis: Findings are deduplicated, cross-checked for consensus, and categorized into Act On, Consider, Noted, and Dismissed buckets with rationale. - Use Case: Before merging a feature branch, run an adversarial review on the full diff to catch correctness bugs and design flaws that a single reviewer would miss, without auto-applying any changes. ## Quick Start Ask the assistant to interrogate the changes on your current branch and stress test the diff against main.

Frequently Asked Questions about interrogate

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a multi-model adversarial code review?▼

Point the skill at a diff, branch, or set of files and it spawns one reviewer per configured model tier with the same prompt and rubric. Findings are then synthesized into a single verdict with consensus and disagreement analysis.

What does the interrogate skill review in a code change?▼

Reviewers apply a rubric covering correctness, root causes versus symptoms, structural integrity, verification, complexity budget, and security. A code-quality lens additionally targets structural simplification, file size growth, and spaghetti branching.

Does the adversarial review automatically apply fixes to my code?▼

No. The deliverable is a synthesized verdict only, and reviewers are explicitly read-only. You decide which findings in the Act On section to address before merging.

How are conflicting findings from different models handled?▼

Findings raised by two or more models independently are treated as high-confidence consensus, while lone-model findings are weighted lower. Disagreements between models are surfaced in an agreement map for context.

What happens if a configured model slug is invalid?▼

The skill checks the valid slugs in the Agent tool error message, picks the closest equivalent Claude tier, and continues the review. It does not block on the slug issue and suggests updating the configured value separately.

Related Skills

Explore More Software Engineering Skills→
TokenRollAITokenRollAI
459

commit

Generate commit messages from git diffs and project history.

Official
Intermediate
pyoclawpyoclaw

code-review-and-quality

Review code changes for correctness, readability, architecture, security, and performance.

Community
Advanced
uzumaki-incuzumaki-inc

health

Run code health checks and compute a weighted 0-10 score.

Official
Advanced
haJ1thaJ1t

cross-model-reviewer

Implement a multi-model code review process using two AI models from different providers.

Community
Advanced