skills.rest
yunshu0909
yunshu0909COMMUNITY
Explore all skills by this author →
GitHub

dual-agent-collaboration

Orchestrates Codex and Claude Code as primary executor and independent peer reviewer across four review gates.

752|107|Updated Jan 19, 2026
Search Tags:#cross-validation#CODEX#dual-agent#review gates#cli orchestration#Code review#Claude code
One-click install
npx skills add https://github.com/yunshu0909/yunshu_skillshub --skill dual-agent-collaboration
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: dual-agent-collaboration
Source: https://github.com/yunshu0909/yunshu_skillshub/tree/main/dual-agent-collaboration
Command: npx skills add https://github.com/yunshu0909/yunshu_skillshub --skill dual-agent-collaboration

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

Single-model coding agents can confidently deliver work that is technically correct but misaligned with what the user actually wanted, with no independent check on requirements, plans, or implementations. This Skill pairs Codex and Claude Code through their local CLIs so one model executes while the other independently challenges intent, plans, code, and final results until a structured ACK verdict is reached.

Core Features & Use Cases

  • Four enforced review gates: Intent, Plan, Implementation, and Final gates run in strict order with persistent state, session continuity rules, and fail-closed BLOCKED verdicts when evidence is missing.
  • Structured cross-model review: A JSON schema enforces verdicts (ACK / NOT_ACK / BLOCKED), P0-P2 findings, artifact inspection proof, and fresh-scan requirements, preventing vague or partial reviews.
  • Read-only peer isolation: The peer CLI runs in read-only sandbox mode with recursion protection, artifact manifest validation, and atomic state files stored outside the repository.
  • Use Case: Before merging a complex cross-module refactor, the primary model sends the diff, test results, and verbatim user request to the peer CLI for a cold implementation review, fixes each P1 finding, resumes the review session, and only proceeds to final acceptance after all gates ACK.

Quick Start

Ask the agent to use the dual-agent-collaboration skill to have the other CLI model independently review this task's requirements, plan, and implementation until every gate returns ACK.

Frequently Asked Questions about dual-agent-collaboration

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

FAQPage Schema
How do I make Codex and Claude Code review each other's work?▼

Run the invoke_peer.py script with --peer auto, a phase (intent, plan, implementation, or final), the project root, and a handoff file. The current model acts as primary and the other CLI reviews read-only, returning a structured ACK, NOT_ACK, or BLOCKED verdict.

What is a multi-agent code review workflow with CLI models?▼

It is a workflow where one model implements while a second model independently reviews requirements, plans, diffs, and test evidence through ordered gates. This Skill enforces gate order, session continuity, and schema-validated verdicts via a wrapper script.

Does the peer reviewer modify my code during review?▼

No. The peer runs strictly read-only: Claude is limited to Read, Grep, and Glob tools with optional OS sandboxing, and Codex runs with read-only sandbox and approval policy set to never. All fixes are made by the primary model.

What happens if the peer CLI is missing or times out?▼

The wrapper returns a BLOCKED verdict with a specific error such as PEER_CLI_NOT_FOUND or PEER_REVIEW_TIMEOUT. The workflow never fabricates a review, and no state file is written until a real structured verdict is obtained.

When should I not use dual-model review?▼

Skip it for low-risk, easily reversible small fixes where four review gates add no value. The Skill's proportionality rule reserves full dual-model collaboration for complex, security-sensitive, or user-requested cross-checking tasks.

Related Skills

Explore More Software Engineering Skills→
wmasfoewmasfoe

deep-interview

Clarify vague user ideas into concrete, executable specifications.

Community
Advanced
bihaviourbihaviour

gh

Automate GitHub pull request creation, merging, and draft issue generation.

Official
Advanced
norrbackanorrbacka

requesting-code-review

Review completed code changes against requirements before merge.

Community
Intermediate
thesyntaxninjathesyntaxninja
2

dispatching-parallel-agents

ImplementingAuthWithErrorHandlingTip:tip:tip:tip:tip:tip:tip:...

Community
Intermediate