harness

Routes feature, bug, and refactor tasks through a 7-stage agent pipeline with verify_all gating.

Updated May 15, 2026
One-click install
npx skills add https://github.com/Alan-IFT/harness-kit --skill harness-alan-ift
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: harness
Source: https://github.com/Alan-IFT/harness-kit/tree/main/skills/harness
Command: npx skills add https://github.com/Alan-IFT/harness-kit --skill harness-alan-ift

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Shipping real code changes with AI often lacks structure: requirements get skipped, designs go unreviewed, and code ships untested. This Skill enforces a disciplined 7-stage pipeline so every non-trivial task passes through requirement analysis, design, gate review, development, code review, QA, and delivery before it is declared done. ## Core Features & Use Cases - Full 7-Stage Pipeline: Routes a task through Requirement Analyst, Solution Architect, Gate Reviewer, Developer, Code Reviewer, QA Tester, and Delivery, each producing a numbered stage document in docs/features/<task-slug>/. - Verify Gate and Archival: A task is not done until .harness/scripts/verify_all passes, after which archive-task harvests insights and moves stage docs to the archive. - Resume and Mode Routing: Skips stages 4-6 when a prior /harness-plan run already produced approved 01-03 documents, and redirects research, design-vetting, or open-ended tasks to the lighter /harness-explore, /harness-plan, and /harness-goal modes. - Use Case: You describe "Add a CSV export button to the orders page" and the pipeline produces requirement, design, review, development, and test documents, then ships verified code. ## Quick Start Run /harness with a clear feature, bug fix, or refactor request such as "fix the race condition in the payment retry" and let the pipeline ship it.

Frequently Asked Questions about harness

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

FAQPage Schema
How do I run a full AI development pipeline for a feature in Claude Code?▼

Invoke /harness with a clear feature, bug, or refactor description. The PM orchestrator dispatches seven stages in order — requirement analysis, design, gate review, development, code review, QA, and delivery — and the task completes only after verify_all passes.

What is the difference between /harness, /harness-plan, /harness-explore, and /harness-goal?▼

/harness runs the full 7-stage pipeline to ship code. /harness-plan stops after design review, /harness-explore produces research findings only, and /harness-goal loops Dev and QA until a measurable criterion is met.

Can I resume a harness pipeline after running /harness-plan first?▼

Yes. If a prior /harness-plan run produced 01-03 documents for the same task slug with an APPROVED FOR DEVELOPMENT verdict, /harness skips stages 4-6 and jumps directly to the Developer stage.

When should I not use the full harness pipeline?▼

Skip it for trivial fixes like typos or single-line edits, which only need a direct edit plus verify_all. Research questions belong in /harness-explore, design vetting in /harness-plan, and open-ended improvement loops in /harness-goal.

Why does a harness task stay incomplete after the code is written?▼

A task is not done until .harness/scripts/verify_all passes and archive-task runs to harvest insights. Declaring completion without the verify gate and archival step violates the pipeline's anti-patterns.