# Godle

> Full version with template examples: https://godle.app/llms-full.txt

> Godle is a free, open AI orchestration standard. It provides 185 expert role contexts, 1,741 prompt templates, 1,731 eval rubrics, 12 multi-agent workflows, and 8 pre-configured team packs — accessible as static JSON (no API key) or via a stateful edge layer (Cloudflare Workers). Godle v3 is agent-native: MCP server and A2A protocol are first-class citizens.

## What Godle Is

Godle is not an AI product you log into. It is a public API and prompt library that developers and AI agents use to build structured, role-aware AI pipelines. Layer 1 is static JSON with CORS enabled (no auth). Layer 2 is a Cloudflare Workers edge layer (API key required) with task execution, sessions, streaming, and eval scoring.

## Key URLs

- [Home](https://godle.app): Overview and prompt generator app
- [AI Prompt Generator](https://godle.app/app): Interactive role-based prompt builder (185 roles, 1741 templates, 20 languages)
- [Jobs — AI Prompts by Role](https://godle.app/jobs): Browse all 185 roles and generate expert prompts for any job function
- [API Documentation](https://godle.app/api): Full v3 API reference with MCP, A2A, sessions, and streaming
- [API Playground](https://godle.app/app/playground/): Interactive sandbox for testing all v3 endpoints live
- [Blog](https://godle.app/blog): Prompt engineering guides and role-specific AI tips

## API v3 — Start Here (Latest)

### Agent Protocol Manifests
- [MCP Manifest](https://godle.app/.well-known/mcp.json): 4 MCP tools — list_roles, get_role, generate_prompt, evaluate_output
- [A2A Agent Card](https://godle.app/.well-known/agent.json): A2A agent identity, capabilities, and task schema
- [Capabilities Index](https://godle.app/api/v3/capabilities.json): Full protocol support, rate limits, SDK version

### Layer 1 — Static JSON (No Auth Required)
- [Manifest](https://godle.app/api/v3/index.json): Version info, stats, all endpoint URLs
- [All Roles](https://godle.app/api/v3/roles.json): 185 expert roles with MCP tool names, A2A task schema, model routing
- [Role Detail](https://godle.app/api/v3/roles/{slug}.json): Full role — templates, inputSchema, outputSchema, successCriteria, memory slots, escalation rules
- [Categories](https://godle.app/api/v3/categories.json): 24 categories with subcategory counts
- [Workflows](https://godle.app/api/v3/workflows.json): 12 multi-step DAG workflow definitions
- [Workflow Detail](https://godle.app/api/v3/workflows/{id}.json): Full workflow with steps, dependencies, approval gates, retry policies
- [Evals](https://godle.app/api/v3/evals.json): Index of 1,731 eval rubrics
- [Eval Detail](https://godle.app/api/v3/evals/{id}.json): Weighted criteria, automated checks, passing score
- [Team Packs](https://godle.app/api/v3/team-packs.json): 8 pre-configured agent team compositions
- [Team Pack Detail](https://godle.app/api/v3/team-packs/{id}.json): Roles, responsibilities, workflows, A2A task schema
- [SDK v3](https://godle.app/api/v3/godle-sdk.js): 7-module UMD SDK — MCP client/server, A2A client/server, plus all v2 methods
- [SDK on GitHub](https://github.com/VC-code-project/godle-sdk): Source code, docs, examples, and manifests (MIT license)

### Layer 2 — Edge Endpoints (API Key Required)
Base URL: https://godle.app/api/v3 (Cloudflare Workers)
- POST /api/v3/tasks — Submit a task; returns taskId immediately
- GET /api/v3/tasks/{id} — Poll task status (submitted | working | completed | failed)
- GET /api/v3/tasks/{id}/stream — SSE stream for real-time task output
- POST /api/v3/sessions — Create a persistent session with role context and memory
- GET /api/v3/sessions/{id} — Restore session (conversation history, memory slots)
- POST /api/v3/eval — Score LLM output against a role's eval rubric

## SDK v3 Modules

The SDK exposes 7 modules in one ~18KB UMD script:
1. **GODLE.mcp.client()** — MCP client: listTools(), callTool(), generateManifest()
2. **GODLE.mcp.server()** — MCP server: addTool(), getManifest(), handleRequest()
3. **GODLE.a2a.client()** — A2A client: sendTask(), getTask()
4. **GODLE.a2a.server()** — A2A server: generateAgentCard(), handle()
5. **GODLE.a2a.toA2ATask()** — Convert GODLE task to A2A wire format
6. **GODLE.a2a.makeMessage()** — Create A2A message object
7. **All v2 methods** — listRoles, getRole, generatePrompt, createWorkflowRun, evaluateOutput, listEvals, listWorkflows, listTeamPacks, getTeamPack, validateInput, getModelHint

## Available Workflows

compliance-audit, content-campaign, data-pipeline-build, design-sprint, feature-development, hiring-pipeline, incident-response, launch-checklist, onboarding-playbook, quarterly-planning, sales-deal-cycle, vendor-evaluation

## Available Team Packs

compliance-team, content-engine, data-team, enterprise-sales-team, growth-team, incident-response-team, product-trio, saas-mvp

## Jobs — 185 AI Role Pages

Each role at godle.app/jobs/{slug} has:
- Expert role context and description
- 6–10 prompt templates (1,741 total across all roles)
- Freeform prompt builder with tone/depth/format controls
- Direct links to ChatGPT, Claude, and Grok
- Available in 20 languages

Top roles: software-engineering, data-science, product-management-core, financial-planning-analysis, content-marketing, enterprise-field-sales, ux-design, hr-business-partnering, corporate-strategy, accounting-close

## API v2 (Stable — Legacy)

Still fully supported. Endpoints at /api/v2/. See /api for full reference.
- [All Roles v2](https://godle.app/api/v2/roles.json)
- [SDK v2](https://godle.app/api/v2/godle-sdk.js)

## API v1 (Legacy — Prompt Library Only)

- [Roles v1](https://godle.app/api/v1/roles.json)
- [SDK v1](https://godle.app/api/v1/godle-sdk.js)

## Key Technical Facts

- No authentication required for Layer 1 (static JSON)
- All Layer 1 endpoints return JSON with CORS: *
- Cache-Control: public, max-age=3600, s-maxage=86400 (Layer 1)
- Layer 2 requires Authorization: Bearer <key> header
- 185 roles across 24 categories, 20 languages
- Model routing hints per template: reasoning | balanced | fast
- Risk levels: low | medium | high
- MCP protocol: 4 tools (list_roles, get_role, generate_prompt, evaluate_output)
- A2A protocol: Google Agent-to-Agent standard, task-based dispatch
- Zero dependencies — vanilla JS, works in browser and Node.js