Skip to content

kiaquila/unicorn-hub

Repository files navigation

Unicorn Hub

Portable multi-agent development blueprint for spec-driven, PR-only, review-gated software work.

This repository is meant to be copied by agents into a new project. It contains the distilled practices, templates, scripts, workflows, and checks needed to bootstrap a multi-agent development system without depending on any private source repository.

Quickstart For Agents

Unicorn Hub can be used in two ways:

  • Analyze and adopt: ask an agent to inspect this blueprint and recommend which parts fit an existing repository.
  • Install: bootstrap the portable workflow files into a target repository, then create or refresh project docs before product-code work.

Use this prompt in the target repository when you want to install the blueprint:

Use the Unicorn Hub repository at https://github.com/kiaquila/unicorn-hub as
the process blueprint. If you do not already have a local copy, clone it
first (the bootstrap script needs a filesystem path). Install the portable
multi-agent development blueprint into the current repository, choose the
closest project profile, adapt placeholders, run local verification, and
prepare a pull request. After bootstrap, follow CREATE-DOCS.md to build
docs_project before creating the first feature spec. Do not copy
project-specific examples or secrets.

If you do not already have a local clone, get one first:

git clone https://github.com/kiaquila/unicorn-hub /tmp/unicorn-hub

Then run, from the target repository, pointing --source at the local Unicorn Hub path:

node /tmp/unicorn-hub/scripts/bootstrap-repo.mjs \
  --source /tmp/unicorn-hub \
  --profile next-app \
  --project-name "Your Project"

--source only accepts a local filesystem path; it does not resolve a Git URL. If the agent says it cannot find Unicorn Hub, share the GitHub URL above so the agent can clone it, then re-run the command with the resulting local path.

Profiles live in profiles/. If no profile fits, use generic values in .unicorn-hub/config.json after bootstrapping.

Current profiles include:

  • generic: fallback for repositories that need manual command adaptation
  • next-app: Next.js applications with TypeScript and hosted previews
  • python-service: Python services with API or worker entrypoints
  • static-vercel: static frontends deployed through Vercel Git integration
  • telegram-bot: Telegram bots with service-style deployment
  • flutter-app: Flutter/Dart apps with mobile targets and optional web demo builds

What Gets Installed

  • Durable documentation system under docs_project/
  • Spec-driven feature memory under specs/<feature-id>/
  • SENAR-style supervised verification fields for goals, scope, acceptance evidence, negative scenarios, and process memory
  • Agent rules: AGENTS.md and CLAUDE.md
  • Local orchestration scripts for worktrees, PR publishing, feature-memory gates, AI review gates, and branch protection
  • GitHub Actions workflows for CI, PR guard, trusted AI review routing, and OSV scanning
  • Supply-chain defaults: pnpm minimumReleaseAge, Dependabot cooldown, pinned package manager, lockfile-oriented installs

Canonical Workflow

  1. Create or refresh project context with the documentation interview.
  2. Create feature memory before product code: spec.md, plan.md, tasks.md.
  3. Implement in one worktree, one branch, one PR.
  4. Run local preflight before every push.
  5. Let CI, PR guard, and AI review fail closed.
  6. Merge only after required checks are green and blocking review findings are resolved.

After Bootstrap

For a new or under-documented target repository, ask the agent to run the installed documentation interview before implementation:

Read CREATE-DOCS.md and ai-docs-guide.md in this repository.
Interview me in small batches and write durable project docs under docs_project/.
When docs are sufficient, create the first specs/<feature-id>/spec.md, plan.md,
and tasks.md. Do not implement product code yet.

If project docs already exist, use the same protocol to review and refresh them instead of duplicating them.

Repository Map

  • docs/ explains the blueprint and operating model.
  • docs/senar-mapping.md maps the SENAR layer onto Unicorn Hub.
  • templates/ contains files copied into target repositories.
  • scripts/ contains reusable Node.js automation.
  • profiles/ describes project-type defaults.
  • tests/ verifies portability, sanitizer rules, and script behavior.

Validation

Run the full local check:

pnpm run preflight

The final check includes a sanitizer pass to ensure the blueprint does not contain secrets, private repository references, owner-specific infrastructure, or source-project residue.

About

Portable multi-agent development blueprint for spec-driven, PR-gated software delivery.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors