HarnessForge

Forge your agent's harness.

Your AI agent builds skills, MCP servers, and plugins. HarnessForge ensures they're built right — so they become a harness worth shipping.

Don't just prompt. Forge.

Status: v0.x — early stage. Conventions are opinionated and may evolve based on community feedback and platform changes.

---

What is this?

HarnessForge is a skill pack — a collection of 5 installable skills that teach your coding agent the engineering discipline of building production-grade agent ecosystem projects.

When your agent activates a HarnessForge skill, it gains access to curated best practices for:

• Structuring projects that agents can discover, understand, and safely invoke
• Writing SKILL.md, AGENTS.md, and MCP tool descriptions that actually work
• Shipping with security, cross-platform compatibility, and proper versioning

Install

Claude Code

# Step 1: Add the marketplace
claude plugin marketplace add OrigArith/harnessforge

# Step 2: Install the plugin
claude plugin install forge@harnessforge

Codex

Public Plugin Directory listing is coming soon. For now, install as a local plugin:

# Step 1: Clone into the Codex plugins directory
git clone https://github.com/OrigArith/harnessforge.git ~/.codex/plugins/harnessforge

# Step 2: Register in your personal marketplace
mkdir -p ~/.agents/plugins
cat > ~/.agents/plugins/marketplace.json << 'EOF'
{
  "name": "local-plugins",
  "interface": { "displayName": "Local Plugins" },
  "plugins": [
    {
      "name": "forge",
      "source": { "source": "local", "path": "./../../.codex/plugins/harnessforge" },
      "policy": { "installation": "AVAILABLE", "authentication": "NONE" },
      "category": "Developer Tools"
    }
  ]
}
EOF

# Step 3: Restart Codex to discover the plugin

See .codex/INSTALL.md for full instructions including repo-scoped install.

Which skill do I need?

Starting a new project?                    → /forge:init
Creating or improving a SKILL.md?          → /forge:skill
Building or debugging an MCP server?       → /forge:mcp
Preparing to release or security auditing? → /forge:ship
Adding Claude Code / Codex support?        → /forge:adapt

Use it

After installation, invoke skills as slash commands:

/forge:init    → Initialize project structure + write AGENTS.md / CLAUDE.md
/forge:skill   → Create SKILL.md files + test trigger accuracy
/forge:mcp     → Develop or debug MCP servers
/forge:ship    → Release readiness + security audit
/forge:adapt   → Add platform adapters

Skills

Skill	Trigger	What it does
forge:init	Creating a new agent ecosystem project, writing AGENTS.md / CLAUDE.md	Project structure (Template A/B/C), root files, directive file authoring
forge:skill	Creating, reviewing, or debugging SKILL.md trigger accuracy	Frontmatter, Progressive Disclosure, token budgets, description trigger testing
forge:mcp	Building or debugging an MCP server	Tool descriptions, schema design, error handling, MCP Inspector diagnostics
forge:ship	Preparing a release or security auditing	54-item checklist, SemVer, OWASP Agentic Top 10, supply chain security
forge:adapt	Adding Claude Code / Codex support	Cross-platform compatibility matrix and adapter templates

Supported Platforms

Platform	Min Version	Status
Claude Code	1.0+	Tested
Codex CLI	0.1+	Tested
OpenCode	0.1+	Compatible, untested (auto-discovers from .claude/skills/ and .agents/skills/)

How it works

HarnessForge follows the Progressive Disclosure model:

1. Tier 1 — Discovery (~100 tokens per skill): Your agent reads skill names and descriptions to decide which skill matches the current task.
2. Tier 2 — Execution (<5,000 tokens per skill): The activated skill's full SKILL.md loads with step-by-step instructions.
3. Tier 3 — On-demand (no limit): Templates, checklists, and examples load from references/ only when needed.

Project Structure

harnessforge/
├── .claude-plugin/      Claude Code plugin manifest + marketplace.json
├── .codex-plugin/       Codex plugin manifest
├── .codex/              Codex local install instructions
├── skills/              5 skill directories (core content)
├── tests/               Smoke tests + trigger eval scaffold
├── config/              Default configuration
├── AGENTS.md            Cross-platform agent directives
├── CLAUDE.md            Claude Code entry shell
├── README.md            This file
├── LICENSE              MIT
├── CONTRIBUTING.md      Contribution guide
├── SECURITY.md          Security policy
└── CHANGELOG.md         Version history

Why "HarnessForge"?

Harness Engineering is the emerging discipline of building runtime frameworks around AI agents — tools, constraints, verification loops, and context delivery systems that turn chaotic AI intelligence into reliable production capability.

Your agent's harness is made of skills, MCP servers, and plugins. HarnessForge is the skill pack that ensures each component is forged with engineering discipline — so the assembled harness actually works.

Contributing

See CONTRIBUTING.md.

License

MIT

---