Moai Meta Harness logo

Moai Meta Harness

OrganizationPopular
modu-ai
moai-meta-harness

DEPRECATED — legacy 7-Phase meta-harness. Redirects to the v4 harness Builder (/moai:harness <natural-language request>) which replaces the static 7-Phase workflow with an orchestrator-direct 4-phase Builder (ANALYZE / PLAN / GENERATE / ACTIVATE) + a manifest-driven dynamic-workflow Runner. Retained as the redirect source for backward-compat invocation paths; the 7-Phase body below is preserved as historical reference, NOT for new harness creation.

Overview

Publishermodu-ai
Repositorymoai-adk
Skill namemoai-meta-harness
Stars
1.2K
Forks
226
Bundled files
4
LicenseApache-2.0
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 4 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by modu-ai on GitHub. Read the source before you install it.

Installation

Install the Moai Meta Harness AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/modu-ai/moai-adk/tree/main/.claude/skills/moai-meta-harness
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/modu-ai/moai-adk.git /tmp/moai-adk
mkdir -p .claude/skills
cp -r /tmp/moai-adk/.claude/skills/moai-meta-harness .claude/skills/moai-meta-harness
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Moai Meta Harness in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Moai Meta Harness on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Moai Meta Harness is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

moai-meta-harness (DEPRECATED — redirect to v4)

DEPRECATION NOTICE — This legacy 7-Phase meta-harness is superseded by the v4 harness Builder (/moai:harness <natural-language request>). The v4 design replaces the static 7-Phase workflow with:

  • An orchestrator-direct 4-phase Builder (ANALYZE -> PLAN -> GENERATE -> ACTIVATE) that holds the plan in Claude's session context and can call AskUserQuestion at the PLAN->GENERATE approval gate (first-class boundary).
  • A manifest-driven dynamic-workflow Runner (hns-<name>-run.js) that reads manifest.json and dispatches specialists per their declared primitive (sub-agent / dynamic-workflow / worktree / /moai goal / adversarial-fan-out) — no heuristic re-derivation.
  • Conditional sub-agent-granular worktree isolation (no mandatory top-level worktree; worktree only for conflict-prone parallel generation).
  • Signal-driven phase synthesis (not a fixed pipeline); the evaluator is conditional (skipped for tasks within the model's solo reliable range).

To create a new harness: issue /moai:harness <natural-language request> (e.g., /moai:harness build a harness for my-project's API development). The orchestrator runs Context-First Discovery on the request, derives a harness <name>, and enters the Builder. Do NOT use the 7-Phase workflow below for new harness creation.

What is preserved here: the 7-Phase body below is retained as the redirect source for backward-compat — so existing invocation paths that reference the legacy workflow land on a deprecation notice + redirect rather than a dead link. The 7-Phase content is historical reference material, NOT active workflow. The revfactory 7-Phase residual grep excludes this body precisely because it IS the redirect source.

Apache 2.0 Attribution: Adapted from revfactory/harness (Apache License 2.0). The 7-Phase workflow below is a MoAI adaptation of the upstream 6-Phase + Evolution Mechanism. See .claude/rules/moai/NOTICE.md for the full third-party notices and the harness policy for derivation history. The v4 Builder (the replacement) is documented in ~/.claude/skills/moai/workflows/harness-builder.md.


v4 Redirect (the active path)

New harness creation: /moai:harness <natural-language request>

This routes to the v4 Builder (orchestrator-direct). The Builder:

  1. Context-First Discovery on the request (extract domain / goal / constraints / scope).
  2. AskUserQuestion Socratic rounds if clarity <100%; derives harness <name> from confirmed intent.
  3. Explicit approval gate (AskUserQuestion, PLAN->GENERATE boundary).
  4. ANALYZE / PLAN / GENERATE / ACTIVATE as orchestrator-direct phases.
  5. Emits 5 artifact types: entry command, Runner Workflow, specialists, companion skills, manifest.json.

Harness lifecycle (list / edit / remove): /moai:harness list|edit|remove <name>.

Harness execution: /harness:<name> (auto-generated thin-wrapper command -> the harness's Runner Workflow).

See harness-builder.md (companion workflow under moai/workflows/) for the full Builder contract and .claude/rules/moai/workflow/dynamic-workflows.md for the Runner's dynamic-workflow primitive.


Legacy 7-Phase Body (HISTORICAL REFERENCE — redirect source, NOT active)

The content below this separator is the original 7-Phase workflow, preserved verbatim as the redirect source. It is NOT the active harness-creation path. New harness creation MUST use /moai:harness <NL request> (v4) above. The revfactory 7-Phase residual grep excludes this body because it IS the legacy source being redirected away from.

Meta-factory skill that architects and generates project-specific agent teams. Adapts the revfactory/harness 7-Phase workflow to MoAI's agent ecosystem. Produces hns-* skills and agent definitions tailored to each project's domain.

Upstream: revfactory/harness (Apache-2.0) — "A meta-skill that designs domain-specific agent teams, defines specialized agents, and generates the skills they use." (2905 stars, 420 forks, created 2026-03-26)

Effectiveness data (design target): +60% avg quality score (49.5 → 79.3), 15/15 win rate, −32% variance (n=15, author-measured A/B, third-party replications pending). Source: Hwang, M. (2026). "Harness: Structured Pre-Configuration for Enhancing LLM Code Agent Output Quality." revfactory/claude-code-harness.


Quick Reference

When to Use

  • /moai project Phase 5+ runs and detects an absent .moai/harness/main.md
  • AGENTS.md contains <!-- moai:harness-start --> markers (installed by the project-harness generation policy, not this skill)
  • User explicitly requests harness generation for their project domain

Key Outputs

ArtifactLocationOwner
Harness config.moai/harness/main.md + extension filesthis skill
Agent definitions.claude/agents/harness/*.mdthis skill
Domain skills.claude/skills/hns-*/SKILL.mdthis skill

All generated artifacts use the hns- prefix — never moai-* (which is template-managed). The moai-harness-* prefix specifically denotes template-managed harness builders (moai-meta-harness, moai-harness-learner) and is NOT used by this generator's emissions.

6 Architectural Patterns (upstream)

Pipeline, Fan-out/Fan-in, Expert Pool, Producer-Reviewer, Supervisor, Hierarchical Delegation.

See phase walkthrough detail for pattern semantics and selection guidance.


Implementation Guide

7-Phase Workflow — Source Mapping

Each MoAI phase maps to upstream revfactory/harness phases (ref: https://github.com/revfactory/harness#workflow):

MoAI PhaseUpstream Harness PhaseOwning AgentInputsOutputs
1. DiscoveryPhase 0 (audit) + Phase 1 domain analysis (Socratic)manager-specUser requestanswers.yaml
2. AnalysisPhase 1 domain analysis (codebase scan)manager-specanswers.yaml + repo stateAnalysis report
3. SynthesisPhase 2 team architecture designmanager-specAnalysis reportSPEC doc with EARS
4. SkeletonPhase 3 agent definition generationmeta-harness (this skill)SPEC doc.moai/harness/main.md + extensions
5. CustomizationPhase 4 skill generationmeta-harness (this skill)Skeleton.claude/agents/harness/*.md + .claude/skills/hns-*/SKILL.md
6. EvaluationPhase 5 integration + Phase 6 validationsync-auditorGenerated artifactsSprint Contract score
7. IterationHarness Evolution Mechanism + Phase 7-5 opsthe harness-learning policy (separate SPEC)Scoring deltasFactory feedback (out of scope)

Phase Summaries

  • Phase 1 (Discovery): manager-spec conducts 16-question Socratic interview (owned by the project-harness generation policy). Output: .moai/harness/answers.yaml
  • Phase 2 (Analysis): manager-spec scans repo (file structure, existing agents/skills, dependency files, test coverage) — strategic analysis is absorbed into manager-spec
  • Phase 3 (Synthesis): manager-spec produces SPEC with EARS requirements selecting one of 6 architectural patterns, defining agent roles, skill categories, acceptance criteria
  • Phase 4 (Skeleton): This skill generates harness skeleton — main.md, agents.md, skills.md extensions, agent definition stubs
  • Phase 5 (Customization): This skill fills the skeleton with domain-specific content referencing the retained MoAI agents (manager-*, builder-harness, sync-auditor) plus per-spawn Agent(general-purpose) domain delegations for domain-specific work
  • Phase 6 (Evaluation): sync-auditor runs Sprint Contract protocol (design constitution §11.5) — 4 dimensions, pass threshold 0.75 (FROZEN floor 0.60)
  • Phase 7 (Iteration): Owned by the harness-learning policy (out of scope for this skill)

See Phase 1-7 detailed walkthrough + agent involvement for full per-phase activity, inputs, outputs, and cross-reference notes.

MoAI Agent Cross-References

This skill orchestrates but does NOT replace existing agents. All named agents are retained MoAI agents — no new agents are introduced; domain-specific work is delegated via per-spawn Agent(general-purpose) with domain instructions. Categories: Planning & Strategy (manager-spec, plan-auditor), Implementation (manager-develop, plus per-spawn Agent(general-purpose) domain delegations), Builders (builder-harness with artifact_type=agent|skill|plugin), Workflow Managers (manager-develop, manager-docs, manager-git; quality gate via /moai gate or the sync-phase-quality-gate.sh Stop hook), Quality (sync-auditor).

See agent cross-references full inventory for per-agent role and phase mapping.

Generated Harness Validation

After Phase 5 (Customization) emits new hns-* skills, this meta-harness automatically hands off to sync-auditor using the Sprint Contract protocol (design constitution §11.5).

4-Dimension Sprint Contract Assessment:

DimensionWhat is Checked
FunctionalityAgent definitions execute their stated purpose; skills have valid trigger conditions
SecurityNo credentials in generated files; tool permissions follow least-privilege
CraftYAML frontmatter valid (CSV allowed-tools, quoted metadata); progressive disclosure configured
ConsistencyDomain alignment with answers.yaml; naming follows the hns-* convention

Scoring:

  • Pass threshold: 0.75 default (configurable via design.yaml pass_threshold)
  • FROZEN floor: 0.60 (design constitution §2, immutable)
  • Scoring rubric: sync-auditor rubric anchoring (design constitution §12, Mechanism 1)

For Phase 3b — Hierarchical Scoring (when harness.yaml sets evaluator_mode: hierarchical), see hierarchical scoring detail.

Design Target Reference: The +60% effectiveness figure from Hwang (2026) — 49.5 → 79.3 in a 15-run A/B study (author-measured, third-party replications pending) — is the design intent for this validation hook. The governing requirement explicitly states this does not require runtime measurement.


Namespace Separation

[HARD] The Skills and Agents namespace splits cleanly into distributed and user-generated halves.

Distributed (template-managed)

The moai-* namespace — every prefix under it (moai-foundation-*, moai-workflow-*, moai-domain-*, moai-ref-*, moai-meta-*, moai-harness-*) — ships with moai-adk. moai update syncs it by removing and reinstalling, so any direct user edit is overwritten on the next update.

Harness assets in this namespace:

  • moai-meta-harness (this skill — 7-Phase generator)
  • moai-harness-learner (project-agnostic lifecycle-management builder)

User-Generated (this meta-harness emits)

The hns-* namespace and the .claude/agents/harness/ directory are user-owned. This meta-harness creates them during the /moai project Phase 5+ interview, tailored to the user's project domain. harness-* and my-harness-* are earlier generations of the same namespace; tooling still recognizes them as user-owned, but new artifacts use hns-*.

User-generated artifacts:

  • .claude/skills/hns-<domain>/SKILL.md — domain-specific skill (e.g., hns-trading, hns-llm-cascade)
  • .claude/agents/harness/<role>.md — agent definition (e.g., .claude/agents/harness/trading-specialist.md)
  • .moai/harness/main.md — harness entry point + extensions

Contract

  • [HARD] This meta-harness MUST emit user-generated skills with the hns- prefix ONLY. Emitting a moai-* (including moai-harness-*) prefixed file during Phase 4 or 5 is a contract violation.
  • [HARD] moai update MUST NOT delete, modify, or sync user-owned skills or .claude/agents/harness/* files. Backup before update is mandatory.
  • [HARD] The template tree MUST NOT contain user-owned skills or .claude/agents/harness/*-specialist.md files. Leak detection triggers a cleanup chore.
  • [HARD] Distinguishing user-owned hns-* / harness-* from the template builder moai-harness-* requires an exact startsWith comparison. A *harness-* substring pattern is prohibited — it matches moai-harness-* too and produces false positives.
  • [HARD] New artifacts carry hns-. Tooling also recognizes the earlier harness- and my-harness- prefixes as user-owned, so harnesses generated before the rename keep working; they are not rewritten automatically.

Generated-Agent Self-Activation Contract

[HARD] Each generated .claude/agents/harness/<role>.md agent MUST be emitted with both of the following frontmatter fields so the generated harness self-activates when the agent is delegated:

  • A skills: frontmatter entry preloading the agent's companion harness-<domain>-* skill. This makes the domain skill load deterministically when the agent runs, rather than relying on auto-discovery which fails silently when the companion skill is absent from the agent's context.
  • A non-empty, trigger-shaped description frontmatter field naming the domain + the observable task-shape, so the orchestrator's .moai/harness/main.md Task-Shape Routing table can dispatch to it.

Both fields are enforced at runtime by the Phase-6 post-generation smoke gate (moai doctor harness, see the project/meta-harness.md workflow Phase 7): a generated agent with an empty description, a dangling skills: reference (pointing at a non-existent hns-* dir), or NO skills: key at all causes the gate to FAIL. A skills:-less agent must not pass silently — that is the auto-discovery failure mode this contract closes. Full emission template + example: project/meta-harness.md § 6.4.1.

Storage Roots

Namespace / PathLocationSourcemoai update behavior
moai-* skills (incl. moai-harness-* builders)~/.claude/skills/moai-*/templateRemoved and reinstalled (overwrite)
hns-* skills (legacy: harness-*, my-harness-*).claude/skills/hns-*/user project (this meta-harness emits — intent declaration)Never deleted or modified; backed up
MoAI agents (retained catalog, FLAT)~/.claude/agents/templateRemoved and reinstalled (overwrite)
Generated harness agents.claude/agents/harness/user project (this meta-harness emits)Never deleted or modified; backed up
Harness config.moai/harness/user projectNever deleted; backed up

Cross-References

  • ~/.claude/skills/moai-meta-harness/SKILL.md § Namespace Separation (this file — canonical generator-side namespace contract)
  • .claude/rules/moai/development/skill-authoring.md § Skills Namespace Policy
  • .claude/rules/moai/development/agent-authoring.md § Agent Directory Convention

Trigger Mechanics

Auto-load Conditions:

  1. /moai project Phase 5+ runs and .moai/harness/main.md is absent
  2. AGENTS.md contains <!-- moai:harness-start --> markers. These markers are installed by the project-harness generation policy during project initialization; this skill does not install them.

Frontmatter Triggers:

This skill loads when any of the following match:

  • Keywords: harness, project-init, meta-skill, agent-team, harness-evolve
  • Agents: manager-spec, sync-auditor
  • Phases: plan, run, sync

Deferred Execution Contract:

This skill provides the workflow recipe and agent cross-references. It does NOT execute /moai project Phase 5+ logic — that invocation is owned by the project-harness generation policy. The separation is intentional:

  • This skill = capability (what to do and how)
  • the project-harness generation policy = invocation wiring (when to do it)

Out of Scope

The following capabilities are explicitly NOT implemented by this skill:

  • 5-layer integration mechanism — owned by the project-harness generation policy. The integration with /moai project phases, hook installation, and AGENTS.md marker management are all delegated to that SPEC.
  • 16-question Socratic interview — owned by the project-harness generation policy. The manager-spec conducts the interview under that SPEC's control.
  • Auto-evolution loop — owned by the harness-learning policy. The learning feedback mechanism (Phase 7) and delta capture are separate work items outside Wave A.
  • Modification of .claude/agents/{moai,harness}/ or static moai-* skills — this meta-harness generates only hns-* prefixed artifacts and has no write access to MoAI's own agent/skill directories.

Works Well With

  • moai-foundation-core — SPEC-First DDD and TRUST 5 quality gates
  • moai-foundation-cc — Claude Code skill/agent authoring standards
  • manager-spec — Conducts Discovery and Synthesis phases
  • sync-auditor — Sprint Contract evaluation in Phase 6
  • builder-harness (artifact_type=agent|skill|plugin) — Artifact generation helper

Upstream: revfactory/harness (Apache-2.0) | MoAI adaptation: the harness policy | v4 supersession: the v4 harness Builder redesign See .claude/rules/moai/NOTICE.md for full Apache 2.0 attribution.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Moai Meta Harness AI skill do?

DEPRECATED — legacy 7-Phase meta-harness. Redirects to the v4 harness Builder (/moai:harness <natural-language request>) which replaces the static 7-Phase workflow with an orchestrator-direct 4-phase Builder (ANALYZE / PLAN / GENERATE / ACTIVATE) + a manifest-driven dynamic-workflow Runner. Retained as the redirect source for backward-compat invocation paths; the 7-Phase body below is preserved as historical reference, NOT for new harness creation.

Why use Moai Meta Harness on TypingMind?

Because you install it once and use it with any model. Moai Meta Harness is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Moai Meta Harness in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/modu-ai/moai-adk/tree/main/.claude/skills/moai-meta-harness. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Moai Meta Harness?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Moai Meta Harness?

As many as you like. As long as a model supports skills, you can use Moai Meta Harness with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Moai Meta Harness AI skill free?

Yes. It is published on GitHub by modu-ai under the Apache-2.0 license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇