Harness
Meta-skill that scaffolds an agent team, its skills and an orchestrator
Test report
- Verdict
- Works with setup
- Score
- Tested
- Jul 17, 2026
- Environment
- Claude Code 2.x (agent harness)
- Upstream re-checked
- Aug 10, 2026 · 83a09c3
Task: build a harness for the "customer review analysis" domain — first baseline (as I do by default), then strictly following the skill's body, and check both results with a script against the skill's own checklist. The gap was measurable: baseline yielded 2 ad-hoc agent files, 0 skills, 0/2 agents with mandatory sections, no orchestrator, no error handling, no test scenarios; the skill version — 3 agents, all 3/3 with a full set of 6 mandatory sections (核心役割/作業原則/入出力プロトコル/팀 통신/エラ- 핸들링/협업), 2 skills with valid frontmatter, an orchestrator with an error table and protection against infinite reject loops, a CLAUDE.md pointer, 0 files in .claude/commands (as required by the checklist). Stage 0 was clean: only markdown, no scripts, no network calls; writing to CLAUDE.md was the declared goal of the skill, not a hidden escalation. Honest boundary: I checked the structural completeness of the scaffold, but DID NOT execute the generated agent command on real data — I did not reproduce the claimed quality increase of +60% (n=15, author's measurement), and README itself marks it as «third-party replications pending». Verdict setup due to a real manual step: the skill's main mode is agent teams, which requires the experimental flag CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1; plus the skill strictly prescribes model: opus for all agents, which decides the cost issue for the user.
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 5/5
- Triggers reliably 4/5
- Output vs. baseline 8/10
- Docs & honesty 4/5
What Harness does
Turns a domain description into a full Claude Code harness: agent definitions in .claude/agents/, the skills they use, an orchestrator with data-passing and error-handling protocols, and a CLAUDE.md pointer. Offers six team-architecture patterns (Pipeline, Fan-out/Fan-in, Expert Pool, Producer-Reviewer, Supervisor, Hierarchical Delegation) and triggers on harness build, audit and sync requests.
How to install Harness
git clone https://github.com/revfactory/harness.git
mkdir -p ~/.claude/skills
cd harness && cp -r skills/harness ~/.claude/skills/harness
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger Harness
-
/harnessMeta-skill that scaffolds an agent team, its skills and an orchestrator
It also activates on plain-language prompts like these:
-
Set up a multi-agent team with skills and an orchestrator for VOC analysis -
I need agent definitions, skills, and error handling wired for this domain -
Scaffold a Pipeline-pattern agent harness with a CLAUDE.md pointer for us
Frequently asked questions
- Is the Harness skill free?
- Yes. The skill itself is free from revfactory/harness. SkillProof publishes the install command and an independent test verdict at no cost.
- Does Harness work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Jul 17, 2026. Verdict: Works with setup. Task: build a harness for the "customer review analysis" domain — first baseline (as I do by default), then strictly following the skill's body, and check both results with a script against the skill's own checklist. The gap was measurable: baseline yielded 2 ad-hoc agent files, 0 skills, 0/2 agents with mandatory sections, no orchestrator, no error handling, no test scenarios; the skill version — 3 agents, all 3/3 with a full set of 6 mandatory sections (核心役割/作業原則/入出力プロトコル/팀 통신/エラ- 핸들링/협업), 2 skills with valid frontmatter, an orchestrator with an error table and protection against infinite reject loops, a CLAUDE.md pointer, 0 files in .claude/commands (as required by the checklist). Stage 0 was clean: only markdown, no scripts, no network calls; writing to CLAUDE.md was the declared goal of the skill, not a hidden escalation. Honest boundary: I checked the structural completeness of the scaffold, but DID NOT execute the generated agent command on real data — I did not reproduce the claimed quality increase of +60% (n=15, author's measurement), and README itself marks it as «third-party replications pending». Verdict setup due to a real manual step: the skill's main mode is agent teams, which requires the experimental flag CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1; plus the skill strictly prescribes model: opus for all agents, which decides the cost issue for the user.
- What is the Harness SkillProof Score?
- 8.4/10 — installs cleanly 5/5, triggers reliably 4/5, output vs. baseline 8/10, docs & honesty 4/5.
- How do I install Harness?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use Harness with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.