skill-creator
Creates SKILL.md files from specifications. Use when user asks to "create a skill", "build a skill", "make a reusable procedure", or "generate skill for [domain]". Produces folder structure, frontmatter, numbered steps, error handling, and validation checks.
SKILL.md
| Name | skill-creator |
| Description | Creates SKILL.md files from specifications. Use when user asks to "create a skill", "build a skill", "make a reusable procedure", or "generate skill for [domain]". Produces folder structure, frontmatter, numbered steps, error handling, and validation checks. |
Brain-Spoke-Flow
<p align="center"> <img src="docs/assets/hub-spoke.png" alt="Hub-and-spoke architecture — Brain orchestrator delegates to five specialist subagents" width="500" /> </p>Brain orchestrator + 5 specialist agents, hub-and-spoke architecture, structured development lifecycle — from interview to committed code.
Brain decomposes requests and delegates. Researcher gathers evidence. Planner builds phased plans. Developer implements. Inspector verifies. Curator maintains the workspace. Context flows down only — agents never talk to each other.
Quick Start
git clone https://github.com/awieczork/brain-spoke-flow.git
cd brain-spoke-flow
python setup/bootstrap.py # creates runtime directories
Open VS Code → Copilot Chat → @brain → describe your task. Brain interviews you, then drives the workflow.
Reset sessions: python setup/reset-sessions.py (--check for dry-run)
Requirements
- VS Code + GitHub Copilot Chat
- Python 3.10+ (stdlib only — no external modules)
How It Works
Every request flows through a phased lifecycle scaled to complexity:
| Scale | Phases | Sessions |
|---|---|---|
| Quick fix | Interview → Development | 1 |
| Focused task | Interview → Research → Planning → Development → Testing → Review → Curation | 1 |
| Feature | Interview → Research → Planning → child Focused tasks | Multi |
| Project | Interview → Research → child Features | Multi |
3 approval gates where you confirm before proceeding: after Interview, after Planning, after Review.
Project Structure
.github/
├── copilot-instructions.md # Always-on project identity
├── agents/core/ # Brain + 5 subagent definitions
├── contracts/ # Delegation & return schemas, session spec
├── hooks/ # 8 deterministic lifecycle hooks (Python)
├── instructions/ # Convention enforcement (glob-scoped)
├── skills/core/ # Creator skills for authoring artifacts
└── prompts/ # Reusable task templates
docs/ # Human docs, diagrams, rendered PNGs
setup/ # bootstrap.py, reset-sessions.py
Runtime data (.github/.sessions/, output/) is gitignored and created by bootstrap.py.
Extending
Add domain-specific artifacts — the framework stays untouched:
| Artifact | Location | Scaffolding skill |
|---|---|---|
| Domain agents | agents/domain/{name}.agent.md | agent-creator |
| Domain skills | skills/domain/{name}/SKILL.md | skill-creator |
| Instructions | instructions/{name}.instructions.md | instruction-creator |
| Prompts | prompts/{name}.prompt.md | — |
Docs
| ARCHITECTURE.md | Full system specification |
| docs/workflow-lifecycle.md | Phase-by-phase lifecycle reference |
| docs/showcase.md | Capabilities overview |
| .github/hooks/scripts/HOOKS.md | Hook technical reference |