Axiom Agent Protocol

"Describe your most painful task. Get a deployment-ready AI agent spec."

A skill that enables any employee to build practical, business-specific AI agents.

Built from the lessons of large-scale AI agent deployments — where most initial attempts fail. Every rule in this skill exists because someone built a bad agent without it.

What It Does

| Phase | Input | Output | |-------|-------|--------| | Phase 1: Recommend | Your role + daily tasks | 10–20 agent recommendations ranked by ROI | | Phase 2: Spec | Your chosen agent | Complete development spec (.md) ready for Claude Code |

Installation

Install the skill into any project in one command:

npx @coe_dev/axiom-agent-protocol

This copies all skill files into .claude/skills/axiom-agent-protocol/ inside your project directory.

Quick Start

Step 1 — Install

cd your-project
npx @coe_dev/axiom-agent-protocol

Step 2 — Open Claude Code and invoke the skill

Either type the slash command directly:

/axiom-agent-protocol

Or just describe your task naturally — Claude Code will activate the skill automatically based on context:

I'm an HR Manager. I spend 3 hours every week screening resumes.
What AI agents can help me?

💡 After installing, restart Claude Code once so it discovers the new skill.

Step 3 — Follow the protocol

The skill will:

1. Ask 4 clarifying questions about your workflow
2. Recommend 10–20 agents ranked by time savings
3. Generate a full development spec for the one you choose
4. Validate the spec scores ≥ 80/100 before delivering it

Repository Structure

axiom-agent-protocol/
├── SKILL.md                          # Main skill (read by agent)
├── README.md                         # This file
├── assets/
│   ├── agent-catalog.json            # 50+ agents organized by department
│   ├── spec-template-simple.md       # Tier-1 spec template
│   ├── spec-template-business.md     # Tier-2 spec template
│   ├── spec-template-autonomous.md   # Tier-3 spec template
│   └── role-profiles.json            # Role-based profiles
├── references/
│   ├── agent-taxonomy.md             # Tier classification system
│   ├── spec-format.md                # How to write excellent specs
│   ├── quality-standards.md          # 100-point scoring rubric
│   ├── anti-patterns.md              # 8 common failure patterns
│   └── workflow.md                   # Full workflow guide
├── scripts/
│   ├── validate-spec.py              # Score a spec file (0–100)
│   └── export-spec.sh                # Export spec to PDF or HTML
├── bin/
│   └── install.js                    # npx install script
└── demos/
    ├── demo-sales-report-agent.md    # Complete Tier-1 example (96/100)
    ├── demo-pr-review-agent.md       # Complete Tier-2 example (94/100)
    └── zendesk-ticket-auto-tagger-assigner-spec.md

Agent Tier Reference

| Tier | Description | Tools | Build Time | |------|-----------|-------|-----------| | Tier 1 · Simple Automation | Single-purpose, predictable, no branching | Claude Cowork, GAS | 4–8 hours | | Tier 2 · Business Automation | Multi-step workflow with decision logic | Claude Code, Devin | 1–3 days | | Tier 3 · Autonomous Execution | Self-directed, 24/7, multi-system | Claude Code, OpenClaw | 1–4 weeks |

Start with Tier 1. Every employee who has successfully built Tier-1 agents is 10x more likely to succeed at Tier 2.

Validate a Spec

Check your spec's quality score before building:

python3 .claude/skills/axiom-agent-protocol/scripts/validate-spec.py my-agent-spec.md

Output:

========================================
Axiom Agent Protocol · Spec Validator
File: my-agent-spec.md
========================================

✅ S1 Agent Name: 10/10
✅ S2 Purpose Clarity: 10/10
⚠️  S3 Trigger Condition: 6/10
   Trigger is vague ('when needed'). Use specific time or event.
...
========================================
TOTAL SCORE: 78/100
VERDICT: ⚠️ NEEDS IMPROVEMENT — Fix before presenting
Sections needing improvement: S3, S7
========================================

The Quality Standard

Every spec delivered by this skill passes a 10-section, 100-point rubric before being shown:

| Section | Points | What it checks | |---------|--------|---------------| | Agent Name | 10 | Follows Verb + Object + Format | | Purpose | 10 | Single sentence with quantified business value | | Trigger | 10 | Concrete time or event | | Input Spec | 10 | Fields, types, sample data | | Output Spec | 10 | Format, destination, sample output | | Scope | 10 | Explicit "does NOT do" list (min 3) | | Examples | 10 | Concrete input/output pairs | | KPIs | 10 | Quantitative Before/After | | Error Handling | 10 | 2+ error types with notification targets | | Security | 10 | Data handling, access control |

Minimum score to present: 80/100.

Updating the Skill

When a new version is published, re-run the install command to get the latest:

npx @coe_dev/axiom-agent-protocol@latest

Why This Exists

When companies roll out AI agent platforms, most initial employee submissions are rated low quality. Not because employees aren't capable — because no one taught them what a good agent spec looks like.

This skill is the structured answer to that problem.

The Mission

Empower every employee to become 10x more effective through AI automation.

Axiom Agent Protocol · v1.1 — Published on npm as @coe_dev/axiom-agent-protocol