Plan. Work. Review. Ship.
Turn Claude Code into a disciplined development partner.
English | 日本語
Claude Code is powerful. Harness turns that raw capability into a delivery loop that is easier to trust and harder to derail.
The 5 verb skills keep setup, plan, work, review, and release on one path. The TypeScript guardrail engine protects execution, and validation can be rerun when you need proof.
What matters here is not the theoretical ceiling of Claude Code. It is what becomes the default operating model once you install a harness.
This is a user-facing workflow snapshot as of 2026-03-06, not a popularity contest. Full notes and source links: docs/github-harness-plugin-benchmark.md
The card below focuses on what becomes the default operating path after install.
Claude Harness is the clearest fit if you want the default path itself to stay planned, guarded, reviewed, and rerunnable.
Supported baseline and latest verified snapshot: see Claude Code Compatibility.
- Claude Code v2.1+ (Install Guide)
- Node.js 18+ (for TypeScript core engine & safety hooks)
# Start Claude Code in your project
claude
# Add the marketplace & install
/plugin marketplace add Chachamaru127/claude-code-harness
/plugin install claude-code-harness@claude-code-harness-marketplace
# Initialize your project
/harness-setupThat's it. Start with /harness-plan.
Don't want to read all this? Just type:
/harness-work all
One command runs the full loop after plan approval. Plan → Parallel Implementation → Review → Commit.
⚠️ Experimental workflow: Once you approve the plan, Claude runs to completion. Validate the success/failure contract in docs/evidence/work-all.md before depending on it in production.
/harness-setupBootstraps project files, rules, and command surfaces so the rest of the loop runs against the same conventions.
/harness-plan"I want a login form with email validation"
Harness creates Plans.md with clear acceptance criteria.
/harness-work # Auto-detect parallelism
/harness-work --parallel 5 # 5 workers simultaneouslyEach worker implements, runs a preflight self-check, and waits for an independent review verdict before completion.
/harness-review
| Perspective | Focus |
|---|---|
| Security | Vulnerabilities, injection, auth |
| Performance | Bottlenecks, memory, scaling |
| Quality | Patterns, naming, maintainability |
| Accessibility | WCAG compliance, screen readers |
/harness-releasePackages the verified result into CHANGELOG, tag, and release handoff steps after implementation and review are complete.
Harness v3 protects your codebase with a TypeScript guardrail engine (core/) — 13 declarative rules (R01–R13), compiled and type-checked:
| Rule | Protected | Action |
|---|---|---|
| R01 | sudo commands |
Deny |
| R02 | .git/, .env, secrets |
Deny write |
| R03 | Shell writes to protected files | Deny |
| R04 | Writes outside project | Ask |
| R05 | rm -rf |
Ask |
| R06 | git push --force |
Deny |
| R07–R09 | Mode-specific and secret-read guards | Context-aware |
| R10 | --no-verify, --no-gpg-sign |
Deny |
| R11 | git reset --hard main/master |
Deny |
| R12 | Direct push to main / master |
Warn |
| R13 | Protected file edits | Warn |
| Post | it.skip, assertion tampering |
Warning |
| Perm | git status, npm test |
Auto-allow |
Runtime differences between Claude Code hooks and Codex CLI gates are documented in docs/hardening-parity.md.
v3 unifies 42 skills into 5 verb skills. Start with the verbs first, then add Breezing, Codex, or 2-agent flows only when you need them.
| Ideas → Plans.md | Parallel implementation | 4-angle code review | Tag + GitHub Release | Project init & config |
| Command | What It Does | Legacy Redirect |
|---|---|---|
/harness-plan |
Ideas → Plans.md |
/plan-with-agent, /planning |
/harness-work |
Parallel implementation | /work, /breezing, /impl |
/harness-work all |
Approved plan → implement → review → commit | /work all |
/harness-review |
4-perspective code review | /harness-review, /verify |
/harness-release |
CHANGELOG, tag, GitHub Release | /release-har, /handoff |
/harness-setup |
Initialize project | /harness-init, /setup |
/memory |
Manage SSOT files | — |
| You Are | Harness Helps You |
|---|---|
| Developer | Ship faster with built-in QA |
| Freelancer | Deliver review reports to clients |
| Indie Hacker | Move fast without breaking things |
| VibeCoder | Build apps with natural language |
| Team Lead | Enforce standards across projects |
claude-code-harness/
├── core/ # TypeScript guardrail engine (strict ESM, NodeNext)
│ └── src/ # guardrails/ state/ engine/
├── skills-v3/ # 5 verb skills (plan/execute/review/release/setup)
├── agents-v3/ # 3 agents (worker/reviewer/scaffolder)
├── hooks/ # Thin shims → core/ engine
├── skills/ # 41 legacy skills (retained for compatibility)
├── agents/ # 11 legacy agents (retained for compatibility)
├── scripts/ # v2 hook scripts (coexist with v3 core)
└── templates/ # Generation templates
Breezing (Agent Teams)
Run entire task lists with autonomous agent teams:
/harness-work breezing all # Plan review + parallel implementation
/harness-work breezing --no-discuss all # Skip plan review, go straight to coding
/harness-work breezing --codex all # Delegate to Codex engine
Phase 0 (Planning Discussion) runs by default—Planner analyzes task quality, Critic challenges the plan, then you approve before coding starts.
| Feature | Description |
|---|---|
| Planning Discussion | Planner + Critic review your plan (default-on) |
| Task Validation (V1–V5) | Scope, ambiguity, overlap, dependency, TDD checks |
| Progressive Batching | 8+ tasks auto-split into manageable batches |
| Hook-driven Signals | Auto-triggers for partial review and next batch |
Cost: ~5.5x tokens (default) vs ~4x (with
--no-discuss). The plan review pays for itself by reducing rework.
Codex Engine
Delegate implementation tasks to OpenAI Codex in parallel:
/harness-work --codex implement these 5 API endpointsCodex implements → Self-reviews → Reports back. Works alongside Claude Code workers.
Setup required: Install Codex CLI and configure API key.
Codex CLI Setup
Use Harness with Codex CLI — no Claude Code required.
Prerequisites: Codex CLI (npm i -g @openai/codex), OpenAI API key (OPENAI_API_KEY), Git.
# 1. Clone the Harness repository
git clone https://github.com/Chachamaru127/claude-code-harness.git
cd claude-code-harness
# 2. Install skills/rules to user scope (~/.codex)
./scripts/setup-codex.sh --user
# 3. Go to your project and start working
cd /path/to/your-project
codexOnce inside Codex, use $harness-plan, $harness-work, $breezing, and $harness-review.
| Flag | Description |
|---|---|
--user |
Install to ~/.codex (shared across projects, default) |
--project |
Install to .codex/ in current directory |
Claude Code users can run
/setup codexinside a session instead.
2-Agent Mode (with Cursor)
Use Cursor as PM, Claude Code as implementer.
/harness-release handoff # Report to Cursor PMPlans.md syncs between both.
Codex Review Integration
Add OpenAI Codex for second opinions:
/harness-review --codex # 4 perspectives + Codex CLICodex selects 4 relevant experts from 16 specialist types via codex exec.
Slide Generation
Generate one-page project intro slides:
/generate-slide- 3 visual patterns (Minimalist / Infographic / Hero)
- 2 candidates per pattern with quality scoring
- Best 3 slides exported to
out/slides/selected/
Dependencies:
GOOGLE_AI_API_KEYand Google AI Studio access.
Video Generation
Generate product videos with JSON Schema-driven pipeline:
/generate-video- JSON Schema as SSOT (Single Source of Truth)
- 3-layer validation: scene → scenario → E2E
- Remotion-based rendering with deterministic output
Dependencies: Requires Remotion project setup and ffmpeg.
Agent Trace
Automatically tracks AI-generated code edits:
.claude/state/agent-trace.jsonl
- Records every Edit/Write operation
- Shows project name, current task, recent edits at session end
- Enables
/sync-statusto compare Plans.md with actual changes
No setup required—enabled by default.
Skill packs can teach a prompt. Harness also enforces behavior at runtime.
- Guardrail engine blocks destructive writes, secret exposure, and force-push patterns on the actual execution path.
- Hooks + review flow keep quality checks close to the tools that edit your repo.
- Validation scripts + evidence pack give you a rerunnable way to confirm docs, packaging, and
/harness-work allbehavior.
| Issue | Solution |
|---|---|
| Command not found | Run /harness-setup first |
harness-* commands missing on Windows |
Update or reinstall the plugin. Public command skills now ship as real directories, so core.symlinks=false no longer hides them. |
| Plugin not loading | Clear cache: rm -rf ~/.claude/plugins/cache/claude-code-harness-marketplace/ and restart |
| Hooks not working | Ensure Node.js 18+ is installed |
For more help, open an issue.
/plugin uninstall claude-code-harnessProject files (Plans.md, SSOT files) remain unchanged.
Harness leverages the latest Claude Code features out of the box.
| Feature | Skill | Purpose |
|---|---|---|
| Agent Memory | harness-work, harness-review | Persistent learning across sessions |
| TeammateIdle/TaskCompleted Hook | breezing | Automated team monitoring |
| Worktree isolation | breezing | Safe parallel writes to the same file |
| HTTP hooks | hooks | JSON POST to Slack, dashboards, metrics |
| Effort levels + ultrathink | harness-work | Auto-injects ultrathink for complex tasks |
| Agent hooks | hooks | LLM-powered code quality guards (secrets, TODO stubs, security) |
${CLAUDE_SKILL_DIR} variable |
all harness-* skills | Stable references to skill-local docs |
agent_id / agent_type fields |
hooks, breezing | Robust teammate identity and role guard |
{"continue": false} teammate response |
breezing | Auto-stop when all assigned tasks are complete |
/reload-plugins |
all harness-* skills | Apply skill/hook edits immediately |
/loop + Cron scheduling |
breezing, harness-work | Active polling with /loop 5m /sync-status |
| PostToolUseFailure hook | hooks | Auto-escalation after 3 consecutive tool failures |
| Background Agent output fix | breezing | Safe run_in_background with output path in completion |
| Compaction image retention | all harness-* skills | Images preserved during context compaction |
| WorktreeCreate/Remove hook | breezing | Worktree lifecycle auto-setup and cleanup |
modelOverrides setting |
harness-setup, breezing | Map model picker aliases to Bedrock, Vertex, or other provider-specific model IDs |
autoMemoryDirectory setting |
session-memory, harness-setup | Store Claude auto-memory in a project-specific path when needed |
CLAUDE_CODE_SESSIONEND_HOOKS_TIMEOUT_MS |
hooks | Give SessionEnd hooks enough time for cleanup and finalize work |
| Full model ID support | agents-v3, breezing | Use claude-sonnet-4-6 style IDs in agent frontmatter and JSON config |
Full list: docs/CLAUDE-feature-table.md
| Resource | Description |
|---|---|
| Changelog | Version history |
| Claude Code Compatibility | Requirements |
| Distribution Scope | Included vs compatibility vs development-only paths |
| Work All Evidence Pack | Success/failure verification contract |
| Cursor Integration | 2-Agent setup |
| Benchmark Rubric | Static vs executed evidence scoring |
| Positioning Notes | Public-facing differentiation language |
| Content Layout | Source docs vs generated outputs convention |
Issues and PRs welcome. See CONTRIBUTING.md.
MIT License — Free to use, modify, commercialize.