A highly structured developer specification and workflow protocol for organizing, executing, and validating autonomous coding tasks dispatched from OpenClaw to Codex.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install codex-dev-collaboration
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install codex-dev-collaboration using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is OpenClaw Codex Collaboration?
The OpenClaw Codex Collaboration Specification defines a reliable protocol designed to orchestrate multi-agent coding workflows safely. This specification establishes a robust communication channel between coordinating controllers and execution agents, providing strict execution guidelines, canary checks, directory standards, and verification steps. By utilizing these standardized guidelines, development teams can eliminate execution-blocking errors, track workspace status systematically, and keep multi-agent software development fully predictable.
Implementing structured protocols through Openclaw Skills guarantees high-performance collaboration while safeguarding local repositories from destructive actions. It creates an explicit barrier for AI agents, dictating strict workspace parameters, dependency guidelines, and output validation loops to secure high-quality code generation.
OpenClaw Codex Collaboration Use Cases
- Establishing a safe, multi-agent automated development pipeline where OpenClaw plans tasks and Codex executes repository refactoring.
- Preventing silent command-line failures in automated DevOps and agent networks through pre-flight execution tests.
- Managing complex workspaces with structured tracking for project architectures, roadmaps, task execution logs, and patch diffs.
- Implementing rigid verification check gates that automatically audit AI modifications against expected contracts before marking tasks as complete.
How OpenClaw Codex Collaboration Works
- Canary Verification Gate: Runs a safe, read-only diagnostic command prior to executing write tasks, verifying CLI parameters, authentication, model configurations, and connectivity.
- Directory Initialization: Sets up standard workspace subdirectories for briefs, architectures, granular tasks, and execution run logs.
- Strict Flag Dispatching: Executes development tasks with precise parameter sequencing, explicitly enforcing top-level CLI arguments like
-a neverahead of the execution command. - Constraint Enforcement: Automatically wraps the developer prompt in a rigid template containing safety limits, file blacklists, write-scope boundaries, and required outputs.
- Post-Execution Zero-Output Audit: Verifies the exit code, non-empty result documentation, JSON event streams, and workspace git status before allowing a success state.
OpenClaw Codex Collaboration Setup
Before executing active development tasks, ensure that your CLI tools are fully verified and authenticated.
# Verify your CLI versions and active help parameters
codex --version
codex exec --help
Run the Canary Self-Check in your repository path to confirm CLI state and argument positioning:
codex -a never exec \
-C "$REPO" \
-s read-only \
--json \
-o "$JOB_DIR/canary-result.md" \
"Just answer: Codex exec canary ok. Do not modify files." \
> "$JOB_DIR/canary.jsonl" 2>&1
echo $? > "$JOB_DIR/canary-exit-code.txt"
Passing Criteria: The exit code in canary-exit-code.txt must be 0, and both canary-result.md and canary.jsonl files must contain content.
OpenClaw Codex Collaboration Data Schema & Taxonomy
Collaboration templates within Openclaw Skills organize project workspaces to ensure consistency across execution cycles.
Project Directory Layout
~/.openclaw-codex/projects/{project_name}/
├── brief.md # Project summary
├── architecture.md # Target system architecture and diagrams
├── roadmap.md # Sequential execution steps
├── tasks/ # Independent task definitions
│ └── 001-task-name.md
└── runs/ # Historical execution attempts
└── 20260618-001/
├── prompt.md # Input task prompt used
├── result.md # Output code analysis/results
├── codex.jsonl # JSON event log outputs
├── diff.patch # Workspace patch diff
├── test.log # Validation test suite outputs
└── status.json # Task state tracking metadata
Bidirectional Handover Schema
Structured contracts are kept in the outputs/collaboration_inbox/ path to manage handoffs between the planning and execution agents:
| Handoff Step | Sender | Receiver | Key Schema Contents |
|---|---|---|---|
| 01 Handoff Contract | OpenClaw | Codex | Endpoint paths, DTO definitions, file whitelists/blacklists, compiler validations, rules. |
| 02 Execution Record | Codex | OpenClaw | Modified files index, validation metrics, deviations, lingering technical issues. |
OpenClaw Codex Collaboration Advanced Features
- Positional CLI Safety Checking: Explicitly enforces argument layout rules, completely avoiding parser halts and silent timeouts caused by misconfigured execution parameters.
- Zero-Output Safety Handlers: Detects consecutive blank iterations, automatically narrowing tasks or pausing workflows to seek human intervention.
- Bidirectional Collaboration Inbox: Embeds isolated directories dedicated to structured API handoffs, validating contract compliance, method signatures, and schema fields.
- Standard Constraint Masking: Blocks unsafe system configurations, unapproved dependencies, destructive Git options, and out-of-scope UI design edits.
SKILL.md
Loading
Related Openclaw Skills
Juejin Article Optimizer
A local CLI optimizer that diagnoses, refines, and formats technical drafts for peak engagement and discoverability on Juejin.
Crisp Grocery
A read-only optimized, secure integration for meal planning, promotion analysis, and basket preparation with the Crisp online supermarket.
AI Engineering Workflow
An engineering-grade workflow for AI coding agents that enforces a strict 5-Gate quality control process, shifting software development from fragile vibe coding to maintainable agentic engineering.
DJTU Vehicle Engineering Thesis Checker & Formatter
A specialized tool that validates DJTU Vehicle Engineering thesis drafts in Markdown and compiles them into beautifully formatted Word documents.
Bounty Hunter
Bounty Hunter is a smart AI agent skill designed to evaluate your technical capabilities and match you with lucrative freelancing, translation, and bug bounty tasks across 10+ platforms.
Skill Policy Enforcer
An automated policy enforcement engine designed to audit agent skills against security, structural, and operational guidelines before installation or publication.
Featured*
