Grand Diomande Research · Full HTML Reader

Clarity Agent Protocol (CAP)

``` ┌─────────────────────────┐ │ CONTRACT REGISTRY │ │ (schemas + traits) │ └───────────┬─────────────┘ │ ┌───────────────────┼───────────────────┐ ↓ ↓ ↓ ┌───────────────┐ ┌───────────────┐ ┌───────────────┐ │ Task Contract │ │ Task Contract │ │ Task Contract │ │ schema: X │ │ schema: Y │ │ schema: Z │ │ traits: [a,b] │ │ traits: [b,c] │ │ traits: [a,c] │ └───────┬───────┘ └───────┬───────┘ └───────┬───────┘ │ │ │ ↓ ↓ ↓ ┌───────────────────────────────────────────────────────┐ │ AGENT DISPATCHER │ │ Matches contra

Agents That Account for Themselves research note experiment writeup candidate score 24 .md

Full Public Reader

Clarity Agent Protocol (CAP)

## What This Is
An agent-agnostic governance protocol for AI agent task execution. Inspired by Stacks/Clarity smart contracts — tasks are structured as contracts with traits, schemas, and validation rules. Any agent, any model, any system can submit work. The protocol validates conformance and rejects non-compliant output with structured feedback.

## Core Concept
Just as Clarity smart contracts on Stacks enforce trait conformance at the chain level, CAP enforces task conformance at the agent orchestration level. Agents don't just "do work" — they fulfill contracts. Contracts define what success looks like. Non-conforming work gets rejected with specific feedback, not vague "try again."

Architecture

                    ┌─────────────────────────┐
                    │   CONTRACT REGISTRY      │
                    │   (schemas + traits)     │
                    └───────────┬─────────────┘
                                │
            ┌───────────────────┼───────────────────┐
            ↓                   ↓                   ↓
    ┌───────────────┐   ┌───────────────┐   ┌───────────────┐
    │  Task Contract │   │  Task Contract │   │  Task Contract │
    │  schema: X     │   │  schema: Y     │   │  schema: Z     │
    │  traits: [a,b] │   │  traits: [b,c] │   │  traits: [a,c] │
    └───────┬───────┘   └───────┬───────┘   └───────┬───────┘
            │                   │                   │
            ↓                   ↓                   ↓
    ┌───────────────────────────────────────────────────────┐
    │                AGENT DISPATCHER                        │
    │  Matches contracts to agents by trait capability       │
    └───────────────────────┬───────────────────────────────┘
                            │
        ┌───────────────────┼───────────────────┐
        ↓                   ↓                   ↓
   Claude Max          Codex CLI          Gemini Pro
   Custom Bot          Human Dev          Any Agent
        │                   │                   │
        ↓                   ↓                   ↓
    ┌───────────────────────────────────────────────────────┐
    │              VALIDATION ENGINE                         │
    │  Checks output against contract schema + traits        │
    │  CONFORM → Accept + merge/deploy                       │
    │  REJECT  → Structured feedback + re-queue              │
    └───────────────────────────────────────────────────────┘

Key Abstractions

### Contracts
A contract defines a unit of work with:
- Schema: Expected input/output structure
- Traits: Capabilities required (`buildable`, `testable`, `deployable`)
- Validation rules: How to check conformance
- Governance: Who can approve, auto-merge thresholds

### Traits (from Clarity)
Composable interfaces agents declare they can fulfill:
- `impl-buildable` — output compiles/builds without errors
- `impl-testable` — includes tests that pass
- `impl-deployable` — ready for production deployment
- `impl-reviewable` — structured for code review
- `impl-documented` — includes documentation
- Custom traits per project

### Governance
- Auto-accept threshold: contracts that score above N
- Review gate: below threshold → human or senior agent review
- Escalation: repeated failures bump to higher-tier agent
- Rejection feedback: structured, actionable — not "try again" but "trait X failed because Y, fix by doing Z"

## Tech Stack
- Runtime: TypeScript/Node.js (portable, runs anywhere)
- Schema: JSON Schema for contract definitions
- Storage: SQLite for local, Supabase for distributed
- Transport: HTTP webhooks + WebSocket for real-time
- CLI: `cap validate`, `cap submit`, `cap status`

Directory Structure

clarity-agent-protocol/
├── CLAUDE.md              # This file
├── README.md              # Public-facing docs
├── package.json
├── tsconfig.json
├── src/
│   ├── index.ts           # Entry point + CLI
│   ├── core/
│   │   ├── contract.ts    # Contract definition + parsing
│   │   ├── trait.ts       # Trait system (impl-trait pattern)
│   │   ├── schema.ts      # JSON Schema validation
│   │   └── governance.ts  # Approval rules, thresholds
│   ├── engine/
│   │   ├── validator.ts   # Validation engine
│   │   ├── dispatcher.ts  # Agent-agnostic dispatch
│   │   ├── rejector.ts    # Structured rejection feedback
│   │   └── escalator.ts   # Auto-escalation logic
│   ├── adapters/
│   │   ├── github.ts      # GitHub webhook adapter
│   │   ├── clawdbot.ts    # Clawdbot session adapter
│   │   ├── claude-code.ts # Claude Code CLI adapter
│   │   ├── codex.ts       # Codex CLI adapter
│   │   └── generic.ts     # Generic HTTP adapter
│   ├── registry/
│   │   ├── contracts/     # Contract definitions (YAML/JSON)
│   │   └── traits/        # Trait definitions
│   └── storage/
│       ├── sqlite.ts      # Local storage
│       └── supabase.ts    # Distributed storage
├── contracts/             # Example contract definitions
│   ├── feature.yaml       # Feature implementation contract
│   ├── bugfix.yaml        # Bug fix contract
│   ├── refactor.yaml      # Refactor contract
│   └── ios-build.yaml     # iOS build + TestFlight contract
└── tests/

Contract Example (YAML)

yaml
name: feature-implementation
version: "1.0"
traits:
  required:
    - buildable
    - testable
    - documented
  optional:
    - deployable
schema:
  input:
    title: string
    description: string
    project: string
    branch: string
  output:
    commits: array
    files_changed: array
    build_status: enum[pass, fail]
    test_results:
      passed: number
      failed: number
      coverage: number
    documentation: string
validation:
  build_status: "must equal 'pass'"
  test_results.failed: "must equal 0"
  test_results.coverage: "must be >= 60"
  files_changed: "must not be empty"
governance:
  auto_accept:
    coverage_threshold: 80
    all_tests_pass: true
    build_clean: true
  escalation:
    max_rejections: 3
    escalate_to: "human-review"
  rejection_feedback:
    format: "structured"
    include_fix_suggestions: true

## Design Principles
1. Agent-agnostic — works with any AI model, any CLI, any API, any human
2. Clarity-inspired — traits, conformance, structured validation (not fuzzy LLM judging)
3. Rejection is data — every rejection teaches the agent what to fix
4. Governance scales — from solo dev to team to enterprise
5. Composable — traits combine, contracts compose, adapters plug in
6. Local-first — runs on a laptop, scales to distributed

Promotion Decision

Attach run IDs, datasets, metrics, and reproduction commands.

Source Anchor

clarity-agent-protocol/CLAUDE.md

Detected Structure

Method · Evaluation · Code Anchors · Architecture