A hybrid spec-driven development framework for Claude Code. Combines Spec Kit's sequential pipeline with BMAD Method's role-based agent specialization.
Instead of one generic AI writing specs, planning, coding, and testing, each phase is handled by a specialized agent with a distinct persona, principles, and domain expertise:
| Phase | Command | Agent | Role |
|---|---|---|---|
| 1 | /specagent.constitution |
Diana | Business Analyst — stakeholder discovery, principles |
| 2 | /specagent.specify |
Marco | Product Manager — PRD, user stories, acceptance criteria |
| 3 | /specagent.plan |
Winston | System Architect — technical design, NFR mapping, trade-offs |
| 4 | /specagent.tasks |
Lucia | Scrum Master — task breakdown, story points, dependencies |
| 5 | /specagent.implement |
Amelia | Senior Developer — TDD, safety protocols, incremental delivery |
| 6 | /specagent.verify |
Rafael | QA Engineer — acceptance verification, scope creep detection |
Each agent only does its job. Marco (PM) never prescribes technical solutions. Winston (Architect) never writes user stories. Amelia (Developer) never redesigns the architecture. Rafael (QA) trusts the spec, not the code.
From Spec Kit, we take:
- The sequential pipeline with gates between phases
- Structured templates with inline agent instructions
- Task format with parallel markers and dependency tracking
- Workflow orchestration files
From BMAD Method, we take:
- Named agent personas with deep identity and communication style
- Activation protocols (load context, adopt persona, greet, execute)
- Customizable agents via
customize.tomlwith 3-tier merge (defaults / team / personal) - Menu-driven sub-actions within each agent
- Safety protocols for implementation (Blast Radius, Reproduction First, Strangler Pattern)
git clone https://github.com/sonoman/spec-agents.git
bash spec-agents/install.sh /path/to/your/projectThis installs:
.claude/skills/specagent.*/— Agent skills (SKILL.md + customize.toml + templates).spec-agents/workflows/— Pipeline orchestration.spec-agents/custom/— Project-specific agent overridesspecs/— Where all specification artifacts live
Open Claude Code in your project:
/specagent.constitution # Diana interviews you about project principles
/specagent.specify # Marco writes a feature PRD from your description
/specagent.plan # Winston designs the technical architecture
/specagent.tasks # Lucia breaks it into ordered, pointed tasks
/specagent.implement # Amelia executes tasks phase by phase with tests
/specagent.verify # Rafael verifies everything against the spec
Each agent reads the previous phase's output. Gates prevent skipping ahead.
Each agent's persona, principles, and behavior can be customized per-project:
# Copy the agent's defaults to your project override
cp .claude/skills/specagent.plan/customize.toml .spec-agents/custom/specagent.plan.toml
# Edit to match your project's needs
# e.g., change principles, add persistent facts, modify menu itemsThe merge order is: skill defaults < project overrides (.spec-agents/custom/).
your-project/
.claude/skills/
specagent.constitution/
SKILL.md # Agent definition
customize.toml # Default configuration
templates/ # Document templates
specagent.specify/
specagent.plan/
specagent.tasks/
specagent.implement/
specagent.verify/
.spec-agents/
workflows/ # Pipeline orchestration
custom/ # Project-specific overrides
specs/
constitution.md
001-feature-name/
spec.md
plan.md
tasks.md
research.md
checklists/
spec-checklist.md
verification-report.md
Pre-built workflow compositions in .spec-agents/workflows/:
| Workflow | Steps | Use when |
|---|---|---|
00-specagent.all.md |
All 6 phases | New project, new feature from scratch |
01-specagent.specify-to-plan.md |
Specify + Plan | Constitution exists, need to design a feature |
02-specagent.plan-to-ship.md |
Tasks + Implement + Verify | Spec and plan approved, ready to build |
- Pipeline structure inspired by GitHub Spec Kit
- Agent roles and activation protocol inspired by BMAD Method
- Skill format adapted from Spec Kit Antigravity Skills