ClaudeMod
Back to browse
Harnesses

Claude Code Harness

Chachamaru127's structured harness for managing multi-file Claude Code projects — task queues, context budgeting, file dependency graphs, and session checkpoints.

Chachamaru127415 starsAdded 1 months ago

Claude Harness

Plan. Work. Review. Ship.
Turn Claude Code into a disciplined development partner.

Latest Release License Claude Code Skills TypeScript Core

English | 日本語


Why Harness?

Claude Code is powerful. Harness turns that raw capability into a delivery loop that is easier to trust and harder to derail.

What changes with Claude Harness: shared plan, runtime guardrails, and rerunnable validation

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.

Compared With Popular Claude Code Harnesses

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.

How the default workflow changes after installing Claude Harness, Superpowers, or cc-sdd

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.


Requirements

  • Claude Code v2.1+ (Install Guide)
  • Node.js 18+ (for TypeScript core engine & safety hooks)

Install in 30 Seconds

# 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-setup

That's it. Start with /harness-plan.


🪄 TL;DR: Verified Work All

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.

/work all pipeline

⚠️ 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.


The 5 Verb Workflow

Plan → Work → Review cycle

0. Setup

/harness-setup

Bootstraps project files, rules, and command surfaces so the rest of the loop runs against the same conventions.

1. Plan

/harness-plan

"I want a login form with email validation"

Harness creates Plans.md with clear acceptance criteria.

2. Work

/harness-work              # Auto-detect parallelism
/harness-work --parallel 5 # 5 workers simultaneously

Each worker implements, self-reviews, and reports.

Parallel workers

3. Review

/harness-review

4-perspective review

PerspectiveFocus
SecurityVulnerabilities, injection, auth
PerformanceBottlenecks, memory, scaling
QualityPatterns, naming, maintainability
AccessibilityWCAG compliance, screen readers

4. Release

/harness-release

Packages the verified result into CHANGELOG, tag, and release handoff steps after implementation and review are complete.


Safety First

Safety Protection System

Harness v3 protects your codebase with a TypeScript guardrail engine (core/) — 13 declarative rules (R01–R13), compiled and type-checked:

RuleProtectedAction
R01sudo commandsDeny
R02.git/, .env, secretsDeny write
R03Shell writes to protected filesDeny
R04Writes outside projectAsk
R05rm -rfAsk
R06git push --forceDeny
R07–R09Mode-specific and secret-read guardsContext-aware
R10--no-verify, --no-gpg-signDeny
R11git reset --hard main/masterDeny
R12Direct push to main / masterWarn
R13Protected file editsWarn
Postit.skip, assertion tamperingWarning
Permgit status, npm testAuto-allow

Runtime differences between Claude Code hooks and Codex CLI gates are documented in docs/hardening-parity.md.


5 Verb Skills, Zero Config

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.

/plan

Ideas → Plans.md

/work

Parallel implementation

/review

4-angle code review

/release

Tag + GitHub Release

/setup

Project init & config

Skills ecosystem

Key Commands

CommandWhat It DoesLegacy Redirect
/harness-planIdeas → Plans.md/plan-with-agent, /planning
/harness-workParallel implementation/work, /breezing, /impl
/harness-work allApproved plan → implement → review → commit/work all
/harness-review4-perspective code review/harness-review, /verify
/harness-releaseCHANGELOG, tag, GitHub Release/release-har, /handoff
/harness-setupInitialize project/harness-init, /setup
/memoryManage SSOT files

Who Is This For?

You AreHarness Helps You
DeveloperShip faster with built-in QA
FreelancerDeliver review reports to clients
Indie HackerMove fast without breaking things
VibeCoderBuild apps with natural language
Team LeadEnforce standards across projects

Architecture

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

Advanced Features

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

Breezing agent teams

Phase 0 (Planning Discussion) runs by default—Planner analyzes task quality, Critic challenges the plan, then you approve before coding starts.

FeatureDescription
Planning DiscussionPlanner + Critic review your plan (default-on)
Task Validation (V1–V5)Scope, ambiguity, overlap, dependency, TDD checks
Progressive Batching8+ tasks auto-split into manageable batches
Hook-driven SignalsAuto-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 endpoints

Codex 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
codex

Once inside Codex, use $harness-plan, $harness-work, $breezing, and $harness-review.

FlagDescription
--userInstall to ~/.codex (shared across projects, default)
--projectInstall to .codex/ in current directory

Claude Code users can run /setup codex inside a session instead.

2-Agent Mode (with Cursor)

Use Cursor as PM, Claude Code as implementer.

/harness-release handoff  # Report to Cursor PM

Plans.md syncs between both.

Codex Review Integration

Add OpenAI Codex for second opinions:

/harness-review --codex  # 4 perspectives + Codex CLI

Codex 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_KEY and 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-status to compare Plans.md with actual changes

No setup required—enabled by default.


Why Harness vs Skill-Pack Only?

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 all behavior.

Troubleshooting

IssueSolution
Command not foundRun /harness-setup first
harness-* commands missing on WindowsUpdate or reinstall the plugin. Public command skills now ship as real directories, so core.symlinks=false no longer hides them.
Plugin not loadingClear cache: rm -rf ~/.claude/plugins/cache/claude-code-harness-marketplace/ and restart
Hooks not workingEnsure Node.js 18+ is installed

For more help, open an issue.


Uninstall

/plugin uninstall claude-code-harness

Project files (Plans.md, SSOT files) remain unchanged.


Claude Code 2.1.74+ Features

Harness leverages the latest Claude Code features out of the box.

FeatureSkillPurpose
Agent Memoryharness-work, harness-reviewPersistent learning across sessions
TeammateIdle/TaskCompleted HookbreezingAutomated team monitoring
Worktree isolationbreezingSafe parallel writes to the same file
HTTP hookshooksJSON POST to Slack, dashboards, metrics
Effort levels + ultrathinkharness-workAuto-injects ultrathink for complex tasks
Agent hookshooksLLM-powered code quality guards (secrets, TODO stubs, security)
${CLAUDE_SKILL_DIR} variableall harness-* skillsStable references to skill-local docs
agent_id / agent_type fieldshooks, breezingRobust teammate identity and role guard
{"continue": false} teammate responsebreezingAuto-stop when all assigned tasks are complete
/reload-pluginsall harness-* skillsApply skill/hook edits immediately
/loop + Cron schedulingbreezing, harness-workActive polling with /loop 5m /sync-status
PostToolUseFailure hookhooksAuto-escalation after 3 consecutive tool failures
Background Agent output fixbreezingSafe run_in_background with output path in completion
Compaction image retentionall harness-* skillsImages preserved during context compaction
WorktreeCreate/Remove hookbreezingWorktree lifecycle auto-setup and cleanup
modelOverrides settingharness-setup, breezingMap model picker aliases to Bedrock, Vertex, or other provider-specific model IDs
autoMemoryDirectory settingsession-memory, harness-setupStore Claude auto-memory in a project-specific path when needed
CLAUDE_CODE_SESSIONEND_HOOKS_TIMEOUT_MShooksGive SessionEnd hooks enough time for cleanup and finalize work
Full model ID supportagents-v3, breezingUse claude-sonnet-4-6 style IDs in agent frontmatter and JSON config

Full list: docs/CLAUDE-feature-table.md


Documentation

ResourceDescription
ChangelogVersion history
Claude Code CompatibilityRequirements
Distribution ScopeIncluded vs compatibility vs development-only paths
Work All Evidence PackSuccess/failure verification contract
Cursor Integration2-Agent setup
Benchmark RubricStatic vs executed evidence scoring
Positioning NotesPublic-facing differentiation language
Content LayoutSource docs vs generated outputs convention

Contributing

Issues and PRs welcome. See CONTRIBUTING.md.


Acknowledgments

  • AI Masao — Hierarchical skill design
  • Beagle — Test tampering prevention patterns

License

MIT License — Free to use, modify, commercialize.

English | 日本語

More Harnesss

Harnesses

claude-simone

Structured project management harness for Claude Code — define milestones and sprints in markdown, then run execution loops that drive Claude through each task.

project-managementmilestonessprints+2
by Helmi
GitHub
Harnesses

Claude Squad

Multi-agent tmux orchestrator — spawn N isolated Claude Code sessions in parallel, each in its own git worktree, then merge the best results back to main.

multi-agentparalleltmux+3
by SMTG AI
GitHub
Harnesses

Claude Swarm

Distributed Claude agent swarms — agents communicate via MCP, share tool access, and collaborate on complex tasks with configurable topology and role specialization.

swarmdistributedmulti-agent+3
by parruda
GitHub

Command Palette

Search for a command to run...