Agent Starter Kit workflow
A scaffold of personas, rules and skills in plain Markdown: Maestro hands work to different models and spares your main plan
- 148 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Maestro launches sub-agents and the CLIs of other providers, passing them project text and prompts through stdin.
- The framework writes memory, session and context files into the project and may create `opencode.json` on boot.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/agent-starter-kitDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Maestro launches sub-agents and the CLIs of other providers, passing them project text and prompts through stdin.
- The framework writes memory, session and context files into the project and may create `opencode.json` on boot.
This is third-party code. Review the repository files before installing.
What it does
Agent Starter Kit is an `.agents/` folder in a project that turns a CLI agent into a small team. There is no installer: the repository is cloned into the project and `AGENTS.md` is linked in the root. Personas, rules and skills are plain Markdown files with no runtime, dependencies or build. The main persona, Maestro, takes a request, breaks it down and hands out work: Architect builds the decision tree and the plan, Coder writes code, Reviewer checks the result, Contextualizer describes the project structure. Each persona has an identity, a playbook, a handoff format and red lines. The `preferredModel` field sets the model: heavy roles can go to cheaper models like DeepSeek, GLM, Kimi and Qwen while the orchestrator stays on your main plan. The framework stores corrections and preferences in long-term memory and keeps session files, so interrupted work resumes where it stopped. There are two modes: quick, where the agent simply reads the `AGENTS.md` rules, and full orchestration, started with the phrase `Please comply with @.agents/ENTRYPOINT.md file.`
Who it is for. Developers paying for a coding-agent plan who want to spread work across several models without installing a separate framework.
Good fit when
- A large task needs a plan, review and memory between sessions, while a single agent loses the thread.
- The Claude Code plan runs out too fast, and coding should go to a cheaper model through OpenCode.
- A team wants a shared set of project rules: `AGENTS.md` is picked up by Claude Code, Codex and OpenCode.
- Review should be done by a different model to catch what the code author considers normal.
- You want a base to fork and extend with your own personas, rules and skills.
Not a fit when
- A quick edit of one file is needed: a plain request in quick mode without orchestration is enough.
- You use one model and do not plan to keep memory and review, so the personas would be extra context weight.
- A ready plugin installed with one command is wanted: this kit is copied into the project by hand.
Example request
Please comply with @.agents/ENTRYPOINT.md file. Add an order export to CSV to the service, plan first.Limitations
The kit is a foundation rather than a finished product, and narrow domain rules are meant to be added in a fork. Full orchestration spends more tokens than quick mode and needs the boot phrase at the start of every session. Each provider needs its own plan or API key and an installed CLI: `claude`, `codex` or `opencode`. The automatic `opencode.json` write with agent bindings works only when `yq` and `jq` are installed; other CLIs need no setup.
How to disable. Delete the `.agents/` folder and the `AGENTS.md` symlink in the project root. If the boot created `opencode.json`, remove it too.
FAQ
How do I connect Agent Starter Kit to Claude Code?
Clone the repository into the `.agents` folder of the project with `git clone git@github.com:ntorga/agent-starter-kit.git .agents` and create the link `ln -s .agents/AGENTS.md AGENTS.md`. Claude Code picks the rules up automatically, and the first message of a session starts full orchestration.
How does quick mode differ from full orchestration?
Quick mode works at once: the agent reads `AGENTS.md` and handles the request alone. Full orchestration starts Maestro, which plans, hands work to personas and keeps memory, at the cost of more tokens.
Can the kit be used with a single model?
Yes. Set `preferredModel` of every persona to your current host, for example `claude`, and everything goes through one provider. Planning, review and memory remain.
How does the kit remember my corrections?
Corrections and preferences go to long-term memory, and unfinished work is tracked in session files. The next launch continues where it stopped.
How do I add my own persona or rule?
Put a Markdown file into `personas/`, `rules/` or `skills/` following the schema in the `README.md` of that folder. Rules to keep out of a fork go into `AGENTS.override.md` in the project root.
Discussion
No questions or reviews yet.
Related
WorkflowsA LangWatch command-line tool that creates the scaffold of a new AI agent project and hands it to a coding assistant
A skill with 25 commands to audit and improve the prompts, tools and agent designs of a project in Cursor, Claude Code and Codex
A bridge between local CLI agents: Codex, Cursor or Claude Code delegate work to Antigravity, Grok, Kimi and seven more.
A multi-agent harness that verifies results by artifacts: stop hooks, an independent judge and an event log for Claude Code, Codex, Cursor and more
An open coding agent for the terminal, browser and desktop that works with models from OpenAI and Claude to Ollama
Agentic Substrate
Agentic Substrate (claude-user-memory)
A short reversible rule set for an agent: research the unclear, plan proportionally, verify the outcome and keep checked knowledge
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Agent Starter Kit (this card) | no | 148 | |
| Superpowers | no | 313,096 | a skill set that gives one agent a development process; this one has personas, rules and work split across different models |
| Maestro: an orchestrator of 39 specialists | no | 486 | an orchestration plugin with 39 specialists for Claude Code, Codex, Gemini CLI and Qwen Code; this one is plain Markdown with no plugin, build or runtime |
README in short
The documentation walks from cloning to the first task, lists the personas and more than twenty skills, from memory and dispatch to security review and web search. The FAQ explains the multi-model design and recommends which model to assign to each role.