Skip to content
This repository was archived by the owner on Jun 11, 2026. It is now read-only.
Public archive

About

Meristem – Knowledge-first agentic app template

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

Meristem logo

😄 We had a lot of fun here — but this repository is now archived. Development has moved to oneticket-core, which continues and expands on everything built here.


Meristem – Knowledge-first Agentic App Template

Meristem is a minimal, knowledge-driven template for building applications with AI agents.

It is not a technical starter — no framework, no boilerplate code. It is a structured environment where agents operate based on explicit product knowledge and reusable skills, from which they can generate and evolve any application.


Why Meristem?

In plants, the meristem is the tissue where growth starts. This repository plays a similar role: a small, structured core of knowledge from which agents can generate and evolve entire applications.


How it works

Meristem runs on GitHub Actions. Specialized agents are invoked by a simple comment on any issue or PR:

/dev add a welcome footer
/po initialize the project board
/qa review this PR
/architect what is the current architecture?

Each agent knows its role, loads the skills it needs, and acts — or responds — directly in the thread. The process emerges from agent interactions. Nothing is hardcoded in a central file.

Agents available out of the box: /agent (generalist), /dev, /qa, /po, /analyst, /architect, /help.


Quick Start

git clone https://github.com/dsissoko/meristem.git my-app
cd my-app

Configure the required secrets in GitHub Settings → Secrets → Actions:

Secret Required Description
OPENCODE_API_KEY ✅ (runner anomalyco) OpenCode API key — opencode.ai/auth
CLI_API_KEY ✅ (runner generic-cli) API key for the active CLI — value depends on cli_name in agents/config.yml
SCRUM_PROJECT_TOKEN ⬜ GitHub PAT with project scope — for Scrum board initialization

Enable GitHub Actions write permissions in GitHub Settings → Actions → General → Workflow permissions:

  • Select Read and write permissions
  • Check Allow GitHub Actions to create and approve pull requests

This is required for agents to create branches, commits, and pull requests.

Open an issue and invoke an agent:

/po I want to build a task manager. React + Primer frontend, mocked backend.

The agent reads AGENTS.md, loads the Meristem process, and guides you from there. No manual setup required.


Structure

.
├── AGENTS.md                        ← agent entry point — process and rules
├── agents/                          ← agent profiles and configuration
│   ├── config.yml                   ← active roles, runner mode, model
│   └── <role>/profile.md            ← role identity, skills, routing rules
├── skills/
│   ├── skills-set.md                ← skill sets per role
│   └── local/                       ← core and project skills
├── .github/workflows/               ← dispatcher, runners, CI/CD
├── .opencode/skills/                ← skill bootstrap for OpenCode
├── .claude/skills/                  ← skill bootstrap for Claude Code
├── .gemini/skills/                  ← skill bootstrap for Gemini
├── .agents/skills/                  ← skill bootstrap for generic agents
└── .openhands/microagents/repo.md   ← OpenHands instructions

After first use, the agent creates business.md, architecture.md, and docs/specs/.


Skills

Skills are reusable instruction sets that agents load on demand. Meristem ships with a curated library covering common needs: GitHub context, PR workflow, product knowledge initialization, Scrum project setup, C4 architecture diagrams, frontend stacks, and more.

Skills are organized in sets by role — /dev loads technical skills, /po loads product skills, /architect loads architecture and C4 diagramming skills. When starting a new project, the agent proposes a matching set based on your stack. You validate before anything is applied.

New skills can be added at any time — from the community or written for your project — without modifying the framework.


What's Next

  • Autonomous mode — agents chain themselves without human validation between steps, implemented and ready to test
  • MCP sandbox enrichment — equip agents with MCP servers (GitHub, documentation search, Playwright) before execution
  • More skill sets — community-contributed sets for Vue, Svelte, Python/FastAPI, mobile, and more
  • Collaborative experiment — a public repo where anyone can open an issue, trigger a free-model agent, and watch their feature land in a PR

Contributions welcome — skills, presets, agent profiles, and workflow improvements. See meristem-test for a live demonstration.

About

Meristem – Knowledge-first agentic app template

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages