Add Provider

Add Provider

Add a new AI coding assistant provider to agentconfig.org's provider comparison system.

Overview

Adding a provider requires coordinated work across 6 parallel work streams:

Total effort: ~2-3 hours with parallelization

When to Use

Use this skill when:

• Integrating a new coding assistant (Cursor, Claude Desktop, Zed with AI, etc.)
• Expanding provider support beyond current offerings
• The provider implements most of the 13 AI primitives
• You want comprehensive comparison data visible to users and AI agents

Prerequisites

Before starting, gather:

• Provider capability audit - Which of the 13 primitives does the provider support?
• File path documentation - Where do config files go (global vs project)?
• Support levels - primitives.ts and comparison.ts share one SupportLevel type: full (native), partial (workarounds), diy (custom setup or not achievable) — one vocabulary, no separate none value

The 13 Primitives

Every provider must map to all 13 primitives across the 8 layers below (layers are not 1:1 with primitives — Instructions has three, Memory & State currently has none); see categories in site/src/data/primitives.ts for the full layer list and scopeModel for the separate where-it-applies vocabulary — scopes are not primitives:

Quick Start

0. 🔍 Research the provider → See RESEARCH-GUIDE.md for capability audit template

   • Visit official documentation
   • Document support level for each of the 13 primitives
   • Verify config file locations
   • Complete this BEFORE writing any code (see pre-implementation checklist in CHECKLIST.md)

1. 📋 Read the detailed process → See PROCESS.md for step-by-step instructions for all 6 streams

2. 📖 Review code examples → See EXAMPLES.md for copy-paste templates for each stream

3. 🎨 Understand patterns → See PATTERNS.md for support levels and naming conventions

4. 🐛 Handle errors → See ERRORS.md for solutions to common issues (including critical generation script updates)

5. ✅ Verify completion → See CHECKLIST.md for verification steps (includes pre-implementation checklist)

6-Stream Workflow at a Glance

Stream 1: Type System (Add provider to union types)
   ↓
Stream 2: Data Layer (Add implementations for all 13 primitives)
   ├→ Stream 3: UI Components (Update comparison table)
   │    ↓
   │  Stream 4: Testing (Update E2E tests)
   │    ↓
   └→ Stream 5: App Integration (Update site copy/docs) [can run in parallel with 3-4]
        ↓
      Stream 6: LLMs Generation (Regenerate machine-readable files)

Parallel execution: Start Stream 5 while Streams 3-4 complete. Stream 1-2 are sequential. Stream 6 must run last.

Key Files to Modify

Example Prompts

Add a new provider from scratch:

Use the add-provider skill to add Cursor as a provider to agentconfig.org.
Research Cursor's implementation of all 13 primitives first, then follow all 6 streams.

Skip to a specific stream:

I've completed Stream 1 (types). Now execute Stream 2 (data layer) to add cursor implementations.

Update existing provider data:

Update Cursor's support level from partial to full for Tool Integrations in primitives.ts.

Success Metrics

✅ Provider added to all type definitions ✅ All 13 primitives have provider implementation data ✅ Comparison table renders with provider column ✅ All E2E tests pass ✅ No TypeScript errors ✅ Production build succeeds ✅ llms-full.txt includes provider data ✅ Responsive design works ✅ Dark mode works

PR Description Best Practices

When opening your pull request, keep it crisp and focused:

What to include:

• Summary: One sentence—what provider, what changed
• Changes: Organized by stream (Types, Data, UI, Tests, Integration, Docs)
• Result: Quick summary of provider's final support coverage
• Testing: Concrete steps to verify (run commands, visit site, click features)
• References: Link to official provider documentation as sources

What to avoid:

• Listing all 13 primitives exhaustively
• Repetitive narrative about each stream
• Verbose technical implementation details

Example: See Cursor provider PR for a reference implementation.

Related Skills

• add-primitive - Add a new AI primitive (expand beyond 11)
• generate-llms - Regenerate llms.txt files
• semantic-commit - Create semantic commit messages

References

For detailed information, see:

• RESEARCH-GUIDE.md - How to research a provider before implementation (capability audit template, decision tree, examples)
• PROCESS.md - Complete step-by-step instructions for all 6 streams
• EXAMPLES.md - Copy-paste code examples for each stream
• PATTERNS.md - Support levels, file locations, naming conventions
• ERRORS.md - Common issues and solutions (including critical generation script updates)
• CHECKLIST.md - Comprehensive verification checklists (includes pre-implementation checklist)