A set of agent skills that let an orchestrating AI agent delegate coding tasks to installed coding CLIs, then review the diff and commit it yourself.
didi/helixharness
A framework for teams to build their own AI coding harness on Codex and Claude Code, carrying project rules and knowledge through the development lifecycle.
- Stars
- 325
- Forks
- 54
- Open issues
- 7
- Last commit
- 3 days ago
What is it?
HelixHarness is a plugin for Codex and Claude Code that brings a team's business knowledge, engineering rules and development workflow into the agent's work. The workflow runs from requirements through Propose, Apply (implementation, tests and review) and Archive, with specs, tasks and evidence kept in the repository.
Its core idea is that project knowledge stays in reviewable repository documents. Stable business rules and implementation lessons are recorded at Archive after the user confirms them, then loaded through indexes when later tasks touch the same area. Persisted workflow state supports resuming unfinished work, and a multi-repo Workspace mode records repository responsibilities across services.
An optional helixharness-platform lets framework maintainers observe usage, compare framework versions on the same cases and evaluate candidate changes before adopting them.
Key features
- 01Propose, Apply and Archive workflow stages with specs and evidence stored in the repository
- 02Project knowledge and engineering constraints organized as docs/knowledge and docs/harness
- 03Verified knowledge recorded at Archive only after user confirmation
- 04Persisted workflow state for resuming unfinished work across sessions
- 05Multi-repo Workspace mode for coordinating work across services
- 06Optional platform for observing, evaluating and iterating on the framework
Use cases
- Teams that want AI coding agents to follow their own engineering rules and architectural boundaries
- Projects where business rules from past features should inform later changes
- Coordinating feature work across several repositories or microservices
- Maintainers who want evidence-based comparisons before changing an agent workflow
Quick start
- 1
Add the Codex plugin marketplace
codex plugin marketplace add https://github.com/didi/helixharness.git - 2
Install the Codex plugin
codex plugin add helix@helixharness-marketplace - 3
In your application repository's agent session, initialize HelixHarness
/helix:init - 4
Run a small feature through Propose, Apply and Archive
/helix:propose task-filter ./requirements/task-filter.md --loop
Things to know
- Currently supports only Codex and Claude Code.
- Requires Python 3.10+, Git and a working, signed-in Codex or Claude Code installation.
In collections
Topics
FAQ
- What is helixharness?
- A framework for teams to build their own AI coding harness on Codex and Claude Code, carrying project rules and knowledge through the development lifecycle. HelixHarness is a plugin for Codex and Claude Code that brings a team's business knowledge, engineering rules and development workflow into the agent's work. The workflow runs from requirements through Propose, Apply (implementation, tests and review) and Archive, with specs, tasks and evidence kept in the repository.
- Is helixharness free to use? What license does it have?
- helixharness is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
- Is helixharness still maintained?
- Actively maintained. The last commit to helixharness was on Nov 4, 2026.
- What can I use helixharness for?
- Teams that want AI coding agents to follow their own engineering rules and architectural boundaries Projects where business rules from past features should inform later changes Coordinating feature work across several repositories or microservices Maintainers who want evidence-based comparisons before changing an agent workflow
- How do I install helixharness?
- Add the Codex plugin marketplace: codex plugin marketplace add https://github.com/didi/helixharness.git Install the Codex plugin: codex plugin add helix@helixharness-marketplace In your application repository's agent session, initialize HelixHarness: /helix:init Run a small feature through Propose, Apply and Archive: /helix:propose task-filter ./requirements/task-filter.md --loop
- Where can I get helixharness?
- The source code and installation instructions are on GitHub: https://github.com/didi/helixharness
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A TypeScript CLI and terminal UI for defining, launching and running persistent teams of Claude Code and Codex agents in tmux.
Installs a spec-driven development workflow as Agent Skills for AI coding agents like Claude Code, Codex, and Cursor, covering requirements, design, tasks, and autonomous implementation.
A local goal-run workspace that gives Codex and Claude Code a charter, task board, receipts, and a verification loop for long coding tasks.
A Chinese-language tutorial collection teaching Claude Code, OpenClaw, Codex and WorkBuddy for AI coding, personal agents and office work, with practice materials.
A coding agent built on pi that routes routine decisions, like permission checks and context pruning, to a small judge model so the main model can focus on code.