English | 한국어 | 日本語

Harness is a team-architecture factory for Claude Code. Say "build a harness for this project" (English) or "하네스 구성해줘" (한국어) or "ハーネスを構成して" (日本語), and the plugin turns your domain description into an agent team and the skills they use — picked from six pre-defined team-architecture patterns.

Harness leverages Claude Code's agent team system to decompose complex tasks into coordinated teams of specialized agents. Say "build a harness for this project" and it automatically generates agent definitions (.claude/agents/) and skills (.claude/skills/) tailored to your domain.

Harness lives at the L3 Meta-Factory layer of the Claude Code ecosystem — the layer that generates other harnesses rather than being one. Inside L3, we pick a specific sub-layer: Team-Architecture Factory.

Archon generates deterministic runtime configurations. Harness generates team architectures (pipeline, fan-out/fan-in, expert pool, producer-reviewer, supervisor, hierarchical delegation) plus the skills agents use. Different sub-layers of the same L3. Pick Archon for runtime determinism, Harness for team architecture, or combine them.

• Agent Team Design — 6 architectural patterns: Pipeline, Fan-out/Fan-in, Expert Pool, Producer-Reviewer, Supervisor, and Hierarchical Delegation
• Skill Generation — Auto-generates skills with Progressive Disclosure for efficient context management
• Orchestration — Inter-agent data passing, error handling, and team coordination protocols
• Validation — Trigger verification, dry-run testing, and with-skill vs without-skill comparison tests

The harness evolution mechanism feeds deltas (what worked / what didn't) back into the factory, so the next generation is measurably better. When a generated harness is used in a real project, the /harness:evolve skill captures the delta between the initial architecture and the shipped one, and feeds it back into the factory so the next generation for a similar domain starts closer to the shipped state.

Initial harness ──▶ Real project use ──▶ Shipped harness
                                              │
                                              ▼ (delta capture via /harness:evolve)
                                        ┌───────────────┐
                                        │  Factory      │◀── better next-gen draft
                                        └───────────────┘

We call this the Harness Evolution Mechanism (KR: 하네스 진화 메커니즘; JA: ハーネス進化メカニズム).

Phase 1: Domain Analysis
    ↓
Phase 2: Team Architecture Design (Agent Teams vs Subagents)
    ↓
Phase 3: Agent Definition Generation (.claude/agents/)
    ↓
Phase 4: Skill Generation (.claude/skills/)
    ↓
Phase 5: Integration & Orchestration
    ↓
Phase 6: Validation & Testing

This fork (hongsw/harness) is the Korean persona-injected variant of revfactory/harness. The marketplace and plugin names differ from upstream so the two coexist — install both side-by-side without conflict.

/plugin marketplace add hongsw/harness
/plugin install harness-korean@harness-korean-marketplace

This installs the harness meta-skill plus korean-persona-search / korean-voice-adapter / korean-persona-harness (3 Korean-flavored skills), Codex CLI compatibility, and verification artifacts.

/plugin marketplace add revfactory/harness
/plugin install harness@harness-marketplace

Both can be active at the same time. If you only want the meta-skill and not the Korean additions, install upstream alone.

# Copy the skills directory to ~/.claude/skills/harness/
cp -r skills/harness ~/.claude/skills/harness

harness/
├── .claude-plugin/
│   └── plugin.json                 # Plugin manifest
├── skills/
│   └── harness/
│       ├── SKILL.md                # Main skill definition (6-Phase workflow)
│       └── references/
│           ├── agent-design-patterns.md   # 6 architectural patterns
│           ├── orchestrator-template.md   # Team/subagent orchestrator templates
│           ├── team-examples.md           # 5 real-world team configurations
│           ├── skill-writing-guide.md     # Skill authoring guide
│           ├── skill-testing-guide.md     # Testing & evaluation methodology
│           └── qa-agent-guide.md          # QA agent integration guide
└── README.md

Trigger in Claude Code with prompts like:

Build a harness for this project
Design an agent team for this domain
Set up a harness

Files generated by Harness:

your-project/
├── .claude/
│   ├── agents/          # Agent definition files
│   │   ├── analyst.md
│   │   ├── builder.md
│   │   └── qa.md
│   └── skills/          # Skill files
│       ├── analyze/
│       │   └── SKILL.md
│       └── build/
│           ├── SKILL.md
│           └── references/

Copy any prompt below into Claude Code after installing Harness:

Deep Research

Build a harness for deep research. I need an agent team that can investigate
any topic from multiple angles — web search, academic sources, community
sentiment — then cross-validate findings and produce a comprehensive report.

Website Development

Build a harness for full-stack website development. The team should handle
design, frontend (React/Next.js), backend (API), and QA testing in a
coordinated pipeline from wireframe to deployment.

Webtoon / Comic Production

Build a harness for webtoon episode production. I need agents for story
writing, character design prompts, panel layout planning, and dialogue
editing. They should review each other's work for style consistency.

YouTube Content Planning

Build a harness for YouTube content creation. The team should research
trending topics, write scripts, optimize titles/tags for SEO, and plan
thumbnail concepts — all coordinated by a supervisor agent.

Code Review & Refactoring

Build a harness for comprehensive code review. I want parallel agents
checking architecture, security vulnerabilities, performance bottlenecks,
and code style — then merging all findings into a single report.

Technical Documentation

Build a harness that generates API documentation from this codebase.
Agents should analyze endpoints, write descriptions, generate usage
examples, and review for completeness.

Data Pipeline Design

Build a harness for designing data pipelines. I need agents for schema
design, ETL logic, data validation rules, and monitoring setup that
delegate sub-tasks hierarchically.

Marketing Campaign

Build a harness for marketing campaign creation. The team should research
the target market, write ad copy, design visual concepts, and set up
A/B test plans with iterative quality review.

Harness is not alone in the Claude Code / agent-framework ecosystem. The following repos live in adjacent layers; each is described in a parallel "X is …, Harness is …" form so you can pick the one that fits your need or combine several.

revfactory/harness-100 — 100 production-ready agent team harnesses across 10 domains, available in both English and Korean (200 packages total). Each harness ships with 4-5 specialist agents, an orchestrator skill, and domain-specific skills — all generated by this plugin. 1,808 markdown files covering content creation, software development, data/AI, business strategy, education, legal, health, and more.

revfactory/claude-code-harness — A controlled experiment across 15 software engineering tasks measuring the impact of structured pre-configuration on LLM code agent output quality.

Key finding: effectiveness scales with task complexity — the harder the task, the greater the improvement (+23.8 Basic, +29.6 Advanced, +36.2 Expert).

Exact phrasing to use everywhere: +60% avg quality (49.5 → 79.3), 15/15 win-rate, −32% variance (n=15, author-measured A/B, third-party replications pending).

Full paper: Hwang, M. (2026). Harness: Structured Pre-Configuration for Enhancing LLM Code Agent Output Quality.

This fork hongsw/harness adds 3 non-invasive skills that inject NVIDIA Nemotron-Personas-Korea (1M rows, CC BY 4.0) into harness-generated agents at runtime — yielding agent definitions whose voice, workplace manner, generational vocabulary and cultural cues actually read as Korean. The personas are data-grounded synthetic (LLM-generated, demographically aligned with real Korean distribution), not actual human responses.

Why this matters — see docs/why-data-grounded-synthetic.md: real-persona-grounded harness is eval substrate differentiation, not just content flavor. It connects directly to recommendation engines like hongsw/clawfit.

Sample outputs — see examples/korean-persona/: preserved baseline-vs-grounded comparisons. Currently 1 scenario (backend developer); contributions of more scenarios welcome.

A blind comparison of the same 5-person standup-meeting task (102 vs 103 lines, same LLM, same workload) showed the persona-injected team produced richer voice differentiation (5 distinguishable speakers vs 5 indistinct), 4 inter-personal exchanges vs 0 (mentoring, gratitude, family-aware on-call negotiation), and Korean-specific manners (단정 회피, 컨펌 톤, 우회 표현) that the generic team did not produce.

Quick path (Claude Code marketplace):

/plugin marketplace add hongsw/harness
/plugin install harness-korean@harness-korean-marketplace

Or one-liner (both runtimes):

git clone https://github.com/hongsw/harness.git && cd harness
./scripts/install-korean-persona.sh --target both       # Claude Code + Codex CLI
pip install huggingface_hub pyarrow
python3 skills/korean-persona-search/scripts/download.py  # Dataset cache, first run only

Once installed, both harnesses coexist. Pick by intent (or invoke explicitly when ambiguous):

The two skills are auto-routed by description keywords — Korean-persona terms (한국어 페르소나, 한국 시나리오, Nemotron Korea) trigger the new skill; everything else falls back to the original harness. State the language explicitly when in doubt.

Compatible with both Claude Code and Codex CLI (same SKILL.md format). Details in docs/quickstart-korean-persona.md and docs/why-data-grounded-synthetic.md. The Korean README (README_KO.md) has the full breakdown including the comparison table and verbatim Korean dialogue examples.

• Agent Teams enabled: CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

Apache 2.0