Portable, self-contained AI agent and skill system for structured software development.
9 role-based agents, 15+ workflow skills, multi-target sync (GitHub Copilot + Claude Code).
- git β for subtree integration
- GitHub CLI (
gh) β required for issue tracking and contributing. AI agents useghto create, update, and close issues. Install via your package manager or add to your devcontainer:{ "features": { "ghcr.io/devcontainers/features/github-cli:1": {} } }
git remote add ai-framework git@github.com:23min/ai-first-framework.git
git subtree add --prefix=.ai ai-framework main --squashbash .ai/scripts/init-project.sh # creates work/ dirs, ROADMAP.md, PROJECT_PATHS.md
bash .ai/scripts/sync-all.sh # syncs to .claude/ and .github/git subtree pull --prefix=.ai ai-framework main --squash
bash .ai/scripts/sync-all.shgit subtree push --prefix=.ai ai-framework feature/my-improvement
# Then open a PR on the framework repoAdd to .devcontainer/devcontainer.json to keep targets synced on rebuild:
{
"postCreateCommand": "bash .ai/scripts/sync-all.sh"
}- Getting Started β How to use the framework
- Visual Overview β Mermaid flowchart of components
- Docs β Full documentation index
Starting a new session? β Use skills/session-start.md - it will guide you to the right agent and skill.
Need specific help?
- Planning an epic β skills/epic-refine.md
- Implementing a milestone β skills/milestone-start.md
- Writing tests β skills/red-green-refactor.md
- Releasing β skills/release.md
- Handling a gap β skills/gap-triage.md
Role-based personas that define focus areas and responsibilities:
- architect - Design decisions, system boundaries, epic planning
- implementer - Coding with minimal risk, following TDD
- tester - Test planning, TDD workflow, regression safety
- documenter - Documentation quality, release notes
- deployer - Infrastructure, packaging, releases
- maintainer - AI framework evolution (agents/skills), repository infrastructure
Reusable workflows for common development tasks:
- Epic lifecycle: epic-refine β epic-start β epic-wrap
- Milestone lifecycle: milestone-draft β milestone-start β milestone-wrap
- Development: red-green-refactor, code-review
- Infrastructure: branching, deployment, release
- Planning: roadmap, gap-triage
- Framework maintenance: framework-review, post-mortem
Global guardrails that apply to every session:
- ALWAYS_DO.md - Core rules, session hygiene, build/test requirements
This framework assumes standard paths. If your project differs, update instructions/PROJECT_PATHS.md.
See instructions/PROJECT_PATHS.md for the full path configuration.
Note: Agent labels show primary responsibility. Many steps involve coordination across multiple agents.
Epic work spans multiple milestones and culminates in a major release:
epic-refine (architect)
β
epic-start (architect)
β
[Plan milestones - architect + documenter]
β
[Execute milestones - see Milestone Work below]
β
epic-wrap (documenter + architect + deployer)
β
[PR or direct merge to main]
β
release (deployer) - Epic release ceremony
Epic-wrap involves:
- Documenter: Archive milestone specs, update roadmaps
- Architect: Verify architecture docs align with implementation
- Deployer: Coordinate merge strategy and release planning
Individual milestones within an epic (or standalone):
milestone-draft (documenter)
β
milestone-start (implementer)
β
red-green-refactor (implementer + tester)
β
code-review (tester)
β
milestone-wrap (documenter)
After milestone-wrap:
- If last milestone in epic β epic-wrap (see Epic Work above)
- If interim milestone β Optional tag/deploy, continue to next milestone
- If standalone milestone β Optional release ceremony
Milestone Release (interim/optional):
milestone-wrap
β
[Optional: Tag milestone version]
β
[Optional: Deploy to staging/preview]
Epic Release (major):
epic-wrap (all milestones complete)
β
[PR or direct merge to main - architect decides]
β
release (deployer) - Version bump, changelog, tag
β
[Deploy to production]
β
[Notify team]
| User Request | Use This Skill |
|---|---|
| "Start a new epic" | epic-refine |
| "Plan milestone X" | milestone-draft |
| "Begin milestone X" / "Continue M-X" | milestone-start |
| "Write tests for..." | red-green-refactor |
| "Review this code" | code-review |
| "Create a branch" | branching |
| "Complete milestone X" | milestone-wrap |
| "Create a release" | release |
| "We found a gap" / "This is missing" | gap-triage |
| "Which task should I start?" | session-start |
Large architectural or product themes that span multiple milestones. Examples:
- Add class-based routing to engine
- Implement simulation service with buffer support
- Build UI performance dashboard
Discrete, shippable units of work with clear acceptance criteria. Examples:
- M-02.10: Add provenance query API
- UI-M-03.05: Build timeline visualization
- SIM-M-01.02: Parse YAML templates
- main - Always green and releasable
- epic/ - Integration branch for multi-milestone epics
- milestone/mX - Integration branch for multi-surface milestones
- feature/-mX/ - Actual work branches
- Milestone spec - Authoritative requirements (stable)
- Tracking doc - Implementation progress (dynamic)
- Keep specs stable, update tracking docs frequently
Core Terms:
- Epic - Large architectural or product initiative spanning multiple milestones
- Milestone - Discrete, shippable unit of work with clear acceptance criteria
- AC - Acceptance Criteria; testable conditions that define milestone completion
- DoD - Definition of Done; checklist that must be satisfied before considering work complete
- TDD - Test-Driven Development; RED (write failing test) β GREEN (implement) β REFACTOR
- Spec - Milestone specification; authoritative requirements document (kept stable)
- Tracking doc - Implementation progress log (updated frequently)
- Session log - Record of decisions, tool usage, and actions during a work session
Workflow Terms:
- Preflight checks - Validations run before starting a skill (e.g., epic context exists)
- Handoff - Transition from one skill/agent to another with context transfer
- Gap - Discovered work not in current milestone scope; triaged for future work
- Post-mortem - Reflection on workflow failures to improve framework
Agent Roles:
- Architect - System design, epic planning, architectural decisions
- Implementer - Coding with minimal risk, following TDD
- Tester - Test planning, validation, regression safety
- Documenter - Documentation quality, release notes
- Deployer - Infrastructure, packaging, releases
- Maintainer - AI framework evolution (agents/skills), repository infrastructure (NOT solution development)
To use this framework in a different project:
- Update ALWAYS_DO.md with your project's conventions
- Customize agent responsibilities for your team structure
- Adjust skill references to match your documentation structure
- Modify trigger phrases for your workflow terminology
- Update release skill for your versioning strategy
Always RED β GREEN β REFACTOR:
- RED: Write failing test first
- GREEN: Implement minimum code to pass
- REFACTOR: Improve structure with tests still passing
Never include hours, days, or effort estimates in milestone docs. Focus on:
- Clear requirements
- Testable acceptance criteria
- Scope boundaries
Work is organized around milestones, not arbitrary sprints or dates:
- Milestones have clear success criteria
- Milestones ship to main when complete
- Next milestone branches from current branch
When milestones or epics complete, documentation must be updated:
- Roadmaps reflect current status
- Architecture docs match reality
- Release notes capture what shipped
π Planning Phase
ββ epic-refine β Clarify scope and decisions
ββ milestone-draft β Write specification
ββ gap-triage β Record and route gaps
ββ branching β Create work branch
π Implementation Phase
ββ milestone-start β Begin work, create tracking
ββ red-green-refactor β TDD loop
ββ code-review β Validate changes
β
Completion Phase
ββ milestone-wrap β Mark complete, update docs
ββ milestone release summary β Capture what shipped
ββ release β Version bump, tag, release notes
ββ epic-wrap β Archive milestone specs together
π§ Framework Maintenance
ββ post-mortem β Learn from workflow failures
ββ framework-review β Evaluate framework effectiveness
β Don't:
- Include time/effort estimates in milestone docs
- Skip writing tests before implementation
- Merge to main without updating documentation
- Use ASCII art diagrams (use Mermaid instead)
- Start coding without a clear milestone spec
- Create branches without checking current branch context
β Do:
- Write tests first (RED phase)
- Keep milestone specs stable, tracking docs dynamic
- Update roadmaps and docs when milestones complete
- Use Mermaid for diagrams
- Confirm epic context before starting milestones
- Follow conventional commit messages
- Unclear which skill to use? β Start with session-start
- Need to plan big work? β Use epic-refine
- Ready to code? β Use milestone-start
- Tests failing? β Review red-green-refactor
- Wrapping up? β Use milestone-wrap
If you want to use this framework in your project and contribute improvements back, use git submodules. Framework changes stay in the framework repo's history β your project only tracks which commit it points to.
git submodule add git@github.com:23min/ai-first-framework.git .aiAdd to .devcontainer/devcontainer.json so the submodule is initialized on rebuild:
{
"postCreateCommand": "git submodule update --init && bash .ai/scripts/sync-all.sh"
}cd .ai && git pull origin main && cd ..
git add .ai
git commit -m "chore: update ai-framework submodule"
bash .ai/scripts/sync-all.shEdit .ai/ files directly in your editor. Changes are committed in the framework repo's own git context:
cd .ai
git checkout -b feature/my-improvement
git add -A && git commit -m "feat: my improvement"
git push origin feature/my-improvement
# Then open a PR on the framework repo
cd ..
git add .ai
git commit -m "chore: update ai-framework submodule"Version: 1.2.0 License: MIT