skills.rest
OpenSourceAGI
OpenSourceAGIOFFICIAL
Explore all skills by this author →
GitHub

deep-review

Reviews pull requests and diffs across multiple dimensions using independent reviewer and verifier agents.

74|11|Updated Jul 4, 2024
Search Tags:#diff analysis#code review#pull request#verification#multi-agent#security review#merge verdict
One-click install
npx skills add https://github.com/OpenSourceAGI/qwksearch-research-agent --skill deep-review-opensourceagi
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: deep-review
Source: https://github.com/OpenSourceAGI/qwksearch-research-agent/tree/main/apps/qwk-in-lobe/.agents/skills/deep-review
Command: npx skills add https://github.com/OpenSourceAGI/qwksearch-research-agent --skill deep-review-opensourceagi

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires zod, vitest, and includes scripts (resource) and references (resource) components.

What problem does it solve? Code reviews done by the same agent that wrote the code tend to rubber-stamp it, and single-pass reviews miss whole categories of defects. This Skill runs multi-dimensional code review with independent reviewers, adversarial verification of every candidate finding, and duplicate consolidation before results reach the report. ## Core Features & Use Cases - Two review modes: Light mode dispatches one independent reviewer against dimension quick checklists for ordinary review asks; Deep mode (explicit /deep-review only) runs the full pipeline of dimension reviewers, per-finding verification, global consolidation, and a structured report. - Fourteen review dimensions: logic, security, performance, release-risk, compatibility, UX, observability, reuse-architecture, code style, and more, each with executable rules and pruning criteria. - Anti-hallucination verification: every candidate finding is falsified by an independent verify subagent returning confirmed, false_positive, or need_more_context verdicts, with severity and likelihood overrides. - Use Case: After finishing a feature branch, ask for a review of the PR. The Skill scopes the diff correctly (three-dot range, submodule diffs included), prunes irrelevant dimensions, dispatches reviewers, verifies findings, and renders a report with a merge verdict, hand-off items for pre-existing issues, and a pre-deploy checklist. ## Quick Start Ask the agent to review the current branch or a specific PR, for example: review PR #123 for problems before merging.

Frequently Asked Questions about deep-review

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

FAQPage Schema
How do I run a multi-dimensional code review on a pull request?▼

Ask the agent to review the PR by URL or number, such as "review PR #123". Light mode runs one independent reviewer by default; invoke /deep-review explicitly for the full pipeline of dimension reviewers, verification, and a structured report with a merge verdict.

What is the difference between light mode and deep mode?▼

Light mode dispatches a single independent reviewer against dimension quick checklists with no verification pass. Deep mode runs parallel dimension reviewers, per-finding adversarial verification, global duplicate consolidation, and a full structured report, and is limited to one run per logical requirement.

Which dimensions does the review cover?▼

Fourteen dimensions including logic, security, performance, release-risk, compatibility, UX, observability, code style, reuse-architecture, and workflow. A pruning table skips dimensions that cannot apply to the diff, such as skipping UX when no user-facing surface changed.

How does the review avoid hallucinated findings?▼

Every candidate finding is independently falsified by a separate verify subagent that reads full context and returns confirmed, false_positive, or need_more_context. Only confirmed findings render in the report, and verifiers can override severity, likelihood, and scope labels.

Can the review handle pre-existing bugs found near the diff?▼

Yes. Findings are labeled introduced or exposed_legacy. Legacy issues are not fixed in the current PR; they render under a hand-off section with culprit attribution from git blame, except triggered P0 bugs which block the merge.

When should deep mode not be used?▼

Do not use deep mode for casual review asks, explain-only questions, or follow-up checks after fixes; those use light mode. Deep mode also does not run in unsupported environments, where the Skill offers light mode instead.

Related Skills

Explore More Software Engineering Skills→
brandonavantbrandonavant
2

implement-github-issue

Automate GitHub issue workflows from reading issues to opening pull requests.

Community
Intermediate
qprjack86qprjack86

azure-deploy

Execute validated Azure application deployments through guarded release workflows.

Community
Advanced
KodingDevKodingDev
5

review

Run automated checks and blind multi-lens reviews to produce a final verdict.

Community
Advanced
rjskenerjskene

classify-issue

Classify GitHub issues and apply pipeline triage labels.

Community
Advanced