HelixHarness plugin
A DiDi plugin for Codex and Claude Code that runs a task from proposal to archive and keeps verified project knowledge in the repo
- 420 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The plugin installs agent hooks, and in the Apply stage it edits code and runs tests and checks in the repository.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/helixharnessDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The plugin installs agent hooks, and in the Apply stage it edits code and runs tests and checks in the repository.
This is third-party code. Review the repository files before installing.
What it does
HelixHarness builds the harness an agent works in from the team's rules, business knowledge and checks. System knowledge lives in `docs/knowledge/`, architectural constraints in `docs/harness/`, task contracts in `docs/specs/`, and checks and gates in `docs/hooks/`. Indexes tell the agent which documents to load for the current step. Work runs as a loop of three commands. `/helix:propose` sets scope and acceptance criteria, `/helix:apply` implements, tests and reviews the code, and `/helix:archive` checks completion and proposes stable rules to keep. They enter shared project knowledge only after a person confirms them. The `--loop` flag runs the whole cycle and stops at decisions and blockers. The plugin remembers unfinished work across sessions and can drive several repositories as one workspace.
Who it is for. Development teams with their own architectural rules and business logic, where the agent starts learning the project from scratch every time.
Good fit when
- The agent keeps breaking the same project rules, and they need to be fixed in the repository.
- A feature touches several services and needs one plan with checks in each repository.
- After each task you want to keep confirmed rules so the next one starts from a ready base.
- Work gets interrupted and must continue in a new session without losing the spec and task state.
Not a fit when
- You work in Cursor, Gemini CLI or another agent: only Codex and Claude Code are supported now.
- The task is small and one-off, so knowledge and spec folders would be overhead.
- You need a process with custom stages and order: configurable stages are only planned.
Example request
/helix:propose task-filter ./requirements/task-filter.md --loopLimitations
Installation needs Python 3.10 or newer, Git and a plugin-capable Codex or Claude Code with working model settings. Review the plugin hooks in the agent's interface before the first run, since their execution depends on host settings. The helixharness-platform for observing and evaluating versions is installed separately and is optional for a regular team. Telemetry stays off until the user turns it on, and no feedback address is set by default.
How to disable. Remove the `helix@helixharness-marketplace` plugin through the agent's plugin manager. The `docs/knowledge/`, `docs/harness/` and `docs/specs/` folders stay in the repository as ordinary documents.
FAQ
How do I install HelixHarness in Claude Code?
Add the marketplace with `claude plugin marketplace add https://github.com/didi/helixharness.git` and install `helix@helixharness-marketplace`. No manual source clone is needed.
How does HelixHarness differ from Superpowers and OpenSpec?
Superpowers offers a set of development practices, and OpenSpec drives changes through specs. HelixHarness adds project constraints and knowledge that grows after each task.
Does `--loop` publish code?
No. It goes through proposal, implementation and archive, stops at decisions and blockers, and never publishes or deploys code.
How many commands does the plugin have?
Nine, all in the `helix` namespace. Besides the three main ones there are `/helix:init`, `/helix:status` for progress and blockers, and `/helix:config` for telemetry and report addresses.
Do I need helixharness-platform?
No, the plugin works without it. The platform is for those who develop the harness itself: it shows usage and compares framework versions.
Discussion
No questions or reviews yet.
Related
PluginsHarness drives a task through plan, work and review: spec and Plans.md live in the repo and a separate reviewer checks the result.
A Claude Code plugin with /readiness and /setup skills that scores a repo for agent work and installs rules, hooks and checks.
ADLC Harness takes a task from intent to PR with one command, and the human approves only four key gates.
A CLI, hooks and skills for long Claude, Codex and Pi work: the issue and PR own the task and review gets a PR with checks
A skill that runs a long task as a package with verifiable steps and on-disk state, so work resumes in a new session.
Codely team skills that run agent work in phases: a plan in a file or GitHub issues, review after each phase and a clean commit
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| HelixHarness (this card) | no | 420 | |
| Superpowers | no | 316,679 | a set of general development practices; here project-specific constraints and knowledge that accumulates after tasks |
| OpenSpec | no | 74,686 | drives changes through specs; here a loop with check gates and a shared team knowledge archive |
README in short
The README tells how the project came from DiDi's fintech team, shows where knowledge and rules live in the repository and gives a refund feature example. Below are installation for two agents, a comparison with neighbours and a description of the evaluation platform.