Sub-Agents Directory
A curated directory of 100+ Claude Code sub-agent prompts and MCP servers, helping you quickly find or generate the right sub-agent for your task.
Provided files contain only README, LICENSE, package., and one sync workflow; no content/ prompt files are included, so per-agent tool grants (tools: Read, Write, Edit, Bash), user confirmation, and sensitive-data handling cannot be verified. sync-content.yml auto-pulls upstream content four times daily and commits with contents:write — an automated external-data write path whose upstream repository is unnamed in the files. No rollback mechanism is documented. Dependencies use caret ranges with no lockfile supplied, so supply-chain security is unconfirmed. No malicious behavior found, but most trust criteria sit at the floor.
README instructions align with package. scripts (bun dev, generate:rules, prisma generate), so self-consistency is adequate; but no lockfile is present and dependencies are loosely ranged, hurting reproducibility. No evidence of failure messaging or error handling in the supplied files.
README clearly addresses Claude Code users and contributors; environment fit (Bun + Next.js 16 + Cloudflare Worker) is explicit. Trigger precision and capability boundaries of individual sub-agents depend on the unseen content/ files — only a frontmatter template is shown.
Information architecture is clear (10 category folders, step-by-step contributing guide); the MIT license file is complete (3); version is 0.1.0 with no CHANGELOG; README declares no known limitations; maintenance is partially evidenced by the daily scheduled sync workflow plus husky/lint, but actual content upkeep depends on an unnamed upstream. No FAQ or complete example files.
As a prompt directory site, output usability rests on described features (categories, search, one-click copy); the contribution flow and sync automation lower upkeep cost, giving real marginal value. But the core value (quality of 100+ prompts) is unverifiable statically, and a full Next.js 16 + Supabase + Prisma + Polar stack is heavy for a directory.
The '100+ sub-agents' claim cannot be checked from the supplied files (content/ absent); content originates from an unnamed upstream repo, precluding cross-source corroboration; the README mixes product description with marketing phrasing without separating fact from inference.
- No actual prompt content is included in the supplied files; the '100+ sub-agents' quantity/quality claim cannot be verified in this static review — inspect content/ before adopting.
- Content is auto-synced from an upstream four times daily and auto-committed, and the upstream repository is unnamed in the provided files — verify upstream identity and review synced prompts for overly broad tool grants (e.g., Bash, Write) before deploying or reusing.
- No lockfile is provided and dependencies use loose caret ranges; assess build reproducibility and dependency supply-chain risk yourself.
- README declares no known limitations and there is no CHANGELOG; the version sits at 0.1.0, indicating limited maturity.
What does this agent do, and when should you use it?
Sub-Agents Directory (repository ayush-that/sub-agents.directory) is a curated collection of sub-agent prompts and MCP servers for Claude Code. It ships as a web directory built with Next.js 16 (App Router) and TypeScript, with content stored as Markdown rule files under content/ and organized into ten categories such as core development, language specialists, infrastructure, quality & security, and data & AI. Users can browse by category, search by keyword, copy prompts with one click, or generate custom sub-agents with a built-in AI rule generator, and can open prompts directly in Claude or ChatGPT. An MCP server registry extends Claude Code with external tools. The project runs on Bun, uses Supabase as its database, is MIT licensed, and accepts community contributions via pull requests.
The directory reads Markdown rule files stored in ten category folders under content/ (01-core-development through 10-research-analysis), each carrying front matter with name, description, and tools fields followed by agent instructions. The site provides a searchable directory, categorized prompts (language specialists, DevOps, security, AI/ML, and more), an MCP server registry, an AI rule generator, one-click copy-to-clipboard for prompts, and open-in-Claude/ChatGPT entry points. To run it, clone the repository, run bun install, start the dev server with bun dev, and view the directory at http://localhost:3000. Contributors add new entries by adding Markdown files in the specified format and submitting a pull request.
- A Claude Code user who needs a ready-made sub-agent prompt for a specific task such as security review, DevOps, or data work.
- A developer who wants to create a custom sub-agent prompt for their workflow using the AI rule generator.
- A team looking to extend Claude Code with external tools and browsing the available MCP server registry.
- A prompt author who wants to contribute their sub-agent prompt to a public directory via a categorized Markdown file and PR.
- A ChatGPT user who wants to open directory prompts directly in their assistant.
What are this agent's strengths and limitations?
- 100+ sub-agent prompts organized into ten clear categories, keeping discovery cheap.
- Includes an MCP server registry to extend Claude Code with external tools, not just a prompt list.
- Offers an AI rule generator plus one-click copy and open-in-Claude/ChatGPT for immediate use.
- Prompts are plain Markdown files with metadata, easy to review, version, and contribute via PRs.
- Core value is tied to the Claude Code ecosystem; usefulness is limited for teams not using it.
- Self-hosting requires Bun, Next.js 16, and Supabase, adding deployment and service dependencies.
- Prompt effectiveness depends on the executing model; the repository itself does not guarantee per-entry quality or maintenance.
- New entries depend on manual PR review, so directory freshness and coverage follow community contribution pace.
How do you install or deploy this agent?
git clone https://github.com/ayush-that/sub-agents.directory.git
cd sub-agents.directory
bun install# Requires Bun (https://bun.sh); the stack uses Supabase as its database.
# Start the development server
bun devThen open http://localhost:3000 to view the directory.
How do you use this agent?
- Open the site at http://localhost:3000 and browse by category or search by keyword. 2. Use one-click copy to put a chosen sub-agent prompt on your clipboard, or use the open-in-Claude/ChatGPT option. 3. For a custom agent, use the site's AI rule generator to create a tailored prompt. 4. When self-hosting, add new entries by placing Markdown files in the matching content/ category folder, following the front-matter format (name, description, tools) plus your instructions.