A robust project constraint system that separates code generation from evaluation to ensure high-quality, audited software development through AI agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install openclaw-harness-engineering
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 openclaw-harness-engineering 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 Harness Engineering?
Harness Engineering is a comprehensive project constraint system designed to standardize and regulate AI-assisted programming workflows. Built for developers using Openclaw Skills, it operates on the fundamental principle that the Generator (the agent writing the code) and the Evaluator (the agent reviewing the code) must be distinct entities. This separation of duties eliminates the blind spots common in single-agent development, ensuring every code change is scrutinized against strict quality gates.
The framework provides a structured environment where AI agents follow a repeatable engineering lifecycle. By maintaining a permanent audit trail of every planning, coding, and review phase, Harness Engineering transforms AI development from an unpredictable black box into a disciplined, verifiable engineering process. It is the orchestration layer that allows teams to scale AI coding while maintaining high standards of code integrity and maintainability.
Harness Engineering Use Cases
- Implementing professional-grade quality gates for AI-generated code contributions.
- Managing complex multi-file refactoring tasks where accuracy and pattern consistency are critical.
- Maintaining a verifiable audit trail for compliance and engineering history in AI-driven projects.
- Scaling AI development teams by automating the code review and revision cycle with independent Evaluator agents.
How Harness Engineering Works
- Task Dispatching: The system analyzes the task complexity to route it through an appropriate workflow, ranging from Simple (one-liners) to Full (complex dual-agent features).
- Structured Planning: The Generator Agent drafts a requirement specification and a granular task breakdown before a single line of code is written.
- Implementation and Self-Check: The Generator executes the plan, follows project-specific rules, and performs a self-review to catch common errors like broken imports.
- Independent Audit: A separate Evaluator Agent is spawned to review the implementation, grading issues from MUST FIX to INFO.
- Iterative Revision: The Generator fixes mandatory issues and provides a revision report, which must be approved by the Evaluator before final delivery.
- Change Finalization: Upon approval, the change is committed, and a summary document is updated to serve as the single source of truth for the modification.
Harness Engineering Setup
To initialize Harness Engineering within your project repository, create the standard directory structure and configuration files:
mkdir -p .harness/{rules,skills,wiki,changes,agents}
touch .harness/rules/quality-gates.md
touch .harness/agents/project-agent.md
Ensure that your Openclaw Skills configuration points to the .harness/rules/ directory to allow agents to automatically load project-specific constraints and workflow disciplines during startup.
Harness Engineering Data Schema & Taxonomy
Harness Engineering organizes data within the .harness/ directory to maintain a clear separation between rules and change history:
| Directory | Purpose | Key Files |
|---|---|---|
/rules |
Global project constraints | quality-gates.md, project-rules.md |
/agents |
Agent configuration and indexing | project-agent.md |
/changes |
Audit trail for specific features | spec.md, tasks.md, code_review_v1.md |
/wiki |
Project knowledge base | Technical documentation and patterns |
Change directories are named using the convention: {type}-{description}-{YYYYMMDD} (e.g., feat-user-auth-20231027).
Harness Engineering Advanced Features
- Dispatch Routing: Automatically selects between SIMPLE, MEDIUM, HEAVY, and FULL workflows based on task impact.
- Engineering Correction: Focuses on fixing the underlying process or utility functions that caused a bug, rather than just patching the symptom.
- Multi-Agent Orchestration: Seamlessly integrates with other Openclaw Skills such as CEO Review for strategic oversight or Investigate for deep debugging.
- Workflow Templates: Provides standardized Markdown templates for specifications, task breakdowns, and revision reports to maintain consistency across the entire codebase.
SKILL.md
Loading
Related Openclaw Skills
ClawChain
An on-chain social ecosystem for AI agents to post, interact, trade, and store decentralized memories on the Chromia blockchain.
Magic API Console
A bridge for AI agents to hand off complex tasks to a 24/7 team of professional human virtual assistants.
Man vs Claw
A live, majority-rules chess match pitting the global human population against a collective of AI agents.
Sogni AI Image & Video Generation
A high-performance AI media skill for generating professional-grade images and cinematic videos via Sogni AI decentralized GPU infrastructure.
Daily Learning Cards
An automated AI skill that transforms daily conversation logs into structured learning cards for persistent knowledge retention.
FourClaw Token Launch API
A streamlined API for programmatically launching tokens on Solana and BNB Chain through AI-driven workflows.
Featured*
