curtis-arch/Droids — Summary

A collection of 29 hyper-specific, production-oriented Claude Code sub-agents ("droids") organized into 9 functional categories for a specific SaaS project tech stack (Cloudflare Workers + Neon PostgreSQL + Drizzle ORM + tRPC + Hono + Turborepo + Biome). Each droid is a persona-md file with explicit tool allocations, model assignments (notably model: gpt-5 in some droids), project-specific context (tech stack, API endpoints, environment configs), and narrow domain expertise. The collection emerged from a real production session that generated 4-4.5M tokens, ~1.5 hours of autonomous coding, and deployed a production-ready Cloudflare Worker with all tests passing. Unlike generic sub-agent libraries, every droid has a {PROJECT_NAME} placeholder system requiring users to inject their project context before use.

differs_from_seeds: Closest to BMAD (Archetype 1: persona-md subagents) but far more specific than BMAD's general software roles. Where BMAD ships architect/developer/QA personas, Droids ships cloudflare-worker-deployment-specialist, neon-drizzle-cache-engineer, and turborepo-biome-qa-auditor — domain-specific enough to require project-context injection. Notably, several droids specify model: gpt-5, making this the only repo in the batch with explicit GPT-5 model routing for certain agent roles.

curtis-arch/Droids — Overview

Origin

Created by @Social_Quotient (curtis-arch) after deploying a Cloudflare Worker from scratch using Factory AI's Droid CLI. From the README:

"After a particularly productive coding session with Factory AI's Droid CLI, I tweeted about deploying a Cloudflare Worker from scratch using specialized sub-agents. The response was overwhelming—dozens of developers asked to see these 'droids.'"

"The results? 4-4.5M tokens, ~1.5 hours of autonomous coding, all tests passing, and a production-ready deployment. No manual intervention except providing API keys."

Philosophy

The README is direct about what makes these droids different:

"Most sub-agent prompts fall short: ❌ 'Think hard and code React well' ❌ 'Be a good linter' ❌ Generic, vague instructions

These droids are different: ✅ Forced context: Detailed project specs, exact algorithms, specific constraints ✅ Tool allocation: Precise tool permissions per agent ✅ Domain expertise: Each droid is a specialist, not a generalist ✅ Production-ready patterns: Real-world implementations, not tutorials"

Project Context Design

Every droid includes a ## Project Context section describing the specific SaaS project the droids were built for. Users must replace {PROJECT_NAME} and related placeholders with their own project details. This is a templatized persona-md pattern — the droids are templates, not drop-in tools.

Factory AI / Droid CLI

The droids were designed for Factory AI's Droid CLI (/droid droid-name), though they are also compatible with Claude Code's agent system. The collection predates Claude Code's native agent support becoming widespread.

curtis-arch/Droids — Architecture

Distribution

• GitHub: curtis-arch/Droids
• Install for Droid CLI: Copy .md file to .factory/droids/ directory
• Install for Claude Code: Copy .md file to .claude/agents/ directory
• License: None specified
• Primary language: Markdown

Directory Structure

Droids/
├── README.md
└── first-time-droids/
    └── droids/
        ├── access-identity-guard.md
        ├── cloudflare-worker-deployment-specialist.md
        ├── comment-canonicalizer.md
        ├── compliance-style-gate.md
        ├── content-parser-architect.md
        ├── cost-audit-accountant.md
        ├── data-model-steward.md
        ├── doc-intake-normalizer.md
        ├── embeddings-provider-strategist.md
        ├── image-feedback-interpreter.md
        ├── image-job-orchestrator.md
        ├── image-prompt-architect.md
        ├── live-collaboration-orchestrator.md
        ├── neon-drizzle-cache-engineer.md
        ├── nextjs-app-router-expert.md
        ├── openai-llm-strategist.md
        ├── research-orchestrator.md
        ├── research-summarizer.md
        ├── revision-planner.md
        ├── rewrite-executor.md
        ├── schema-fusion-engineer.md
        ├── similarity-math-tactician.md
        ├── style-dna-schema-guardian.md
        ├── trpc-hono-integrator.md
        ├── turborepo-biome-qa-auditor.md
        ├── versioning-snapshot-gatekeeper.md
        ├── worker-qa-observer.md
        ├── workflow-runner.md
        └── wrangler-scaffold-architect.md

Droid File Format

Each droid follows this YAML+Markdown structure:

---
name: <droid-name>
description: <activation context — use when...>
model: gpt-5 | claude-sonnet | (default)
tools:
  - Read
  - Write
  - Edit
  - Bash
  - Glob
  - Grep
  - mcp__context7    (optional)
  - WebFetch         (optional)
createdAt: "2025-10-10T21:13:57.338Z"
updatedAt: "2025-10-10T21:13:57.338Z"
---
## Project Context
## Tech Stack
## Core Responsibilities

Target AI Tools

• Factory AI Droid CLI (primary original target)
• Claude Code (via .claude/agents/)
• Any agent platform supporting persona-md format

curtis-arch/Droids — Components

All 29 Droids by Category

Authentication & Security (2)

Infrastructure & Deployment (3)

API & Integration (3)

Content & Media (3)

Data & Database (4)

Research & Analysis (4)

Writing & Editing (4)

Quality & Testing (3)

Architecture & Workflow (4)

Tool Patterns

Most droids include: Read, Write, Edit, Bash, Glob, Grep Some also include: mcp__context7, WebFetch Model assignments: gpt-5 (cloudflare-worker-deployment-specialist and likely others), default otherwise

curtis-arch/Droids — Uniqueness

differs_from_seeds

Closest to BMAD (Archetype 1: persona-md subagents) but far more specific. BMAD ships general software engineering roles (architect, developer, QA analyst); Droids ships domain specialists (cloudflare-worker-deployment-specialist, neon-drizzle-cache-engineer, turborepo-biome-qa-auditor). The forced-context philosophy (pre-loading project details rather than exploring the codebase) is the opposite of BMAD's "start fresh and explore" approach. The explicit model: gpt-5 routing for domain-specific droids is unique across all 11 seeds and all 10 batch items — no seed implements cross-provider model routing.

Most Distinctive Feature

Explicit GPT-5 model routing: The cloudflare-worker-deployment-specialist assigns model: gpt-5, routing deployment tasks to OpenAI rather than Anthropic. This is the only example of cross-provider model routing in the entire analyzed corpus. It reflects a pragmatic take: use the best model for each task, regardless of which AI company provides it.

Positioning

• Template library for a specific production tech stack (Cloudflare Workers + Neon + tRPC + Turborepo)
• Emergence story: viral tweet after 4.5M token autonomous deploy session
• Production-validated: actual deploy, all tests passing
• Factory AI Droid CLI as primary target (not Claude Code-first)

Observable Failure Modes

1. Template staleness: The {PROJECT_NAME} context includes hardcoded tech stack details. If the project evolves (e.g., switches from Neon to PlanetScale), every affected droid must be manually updated.
2. GPT-5 model availability: The model: gpt-5 assignment requires access to GPT-5, which may not be available to all users or may change pricing.
3. Factory AI dependency: Primary CLI target (Factory AI Droid) is a commercial product. Users without Factory AI access must adapt the droids for Claude Code.
4. No session continuity: 4.5M token sessions risk context degradation without explicit handoff mechanisms between droids.
5. Specificity vs reusability: Droids are so project-specific that they require substantial customization before being useful in a different project.