SpecSafe

SpecSafe is a two-phase software engineering framework for AI-assisted development. It provides structured planning workflows to reduce ambiguity, then enforces test-driven implementation through small spec slices — regardless of which AI coding tool you use.

For the full canonical workflow, see docs/SPECSAFE-CANONICAL-WORKFLOW.md. For the proposed cross-product assurance seam and future operating plan, see SpecSafe Evidence Chain and Readiness Extension.

Within the governed delivery architecture, SpecSafe remains repo-local assurance: Telar owns business intent, policy, stable identity, human authority, and the normalized ledger; Ultimate Harness is the sole Run Control; OMP Pantheon is the OMP Execution Adapter.

Two-Phase Workflow

PHASE 1: PLANNING
brainstorm → principles → brief → PRD → UX → architecture → readiness

PHASE 2: DEVELOPMENT
spec slice → tests → implementation → verify/QA → complete (or loop back)

Phase 1: Planning

Planning removes ambiguity before implementation begins. Each planning skill produces an artifact that informs the next stage.

Step	Skill	Description
1	specsafe-brainstorm	Divergent exploration of possibilities
2	specsafe-principles	Product principles, non-goals, quality priorities
3	specsafe-brief	Concise product/business framing document
4	specsafe-prd	Testable requirements with user journeys and acceptance criteria
5	specsafe-ux	UX design: tokens, components, flows, accessibility
6	specsafe-architecture	System architecture with ADRs and technology decisions
7	specsafe-readiness	Pre-development coherence check: GO / NEEDS REVISION / BLOCKED

Phase 2: Development

Development begins only after the change is understood well enough to be implemented as a small spec slice.

Stage	Skill	Persona	What happens
SPEC	specsafe-new, specsafe-spec	Kai (Mason)	Create and refine a spec with requirements, acceptance criteria, scenarios
TEST	specsafe-test	Reva (Forge)	Generate test files from spec scenarios — all tests start skipped
CODE	specsafe-code	Zane (Bolt)	TDD red-green-refactor: unskip one test, write code to pass, repeat
VERIFY	specsafe-verify	Lyra (Warden)	Run tests, validate against spec, loop back on failure
QA	specsafe-qa, specsafe-complete	Lyra (Warden), Cass (Herald)	Full QA report with GO/NO-GO, then human approval gate

Quick Start

# Install globally
npm install -g @specsafe/cli

# Initialize a project
specsafe init my-project

# Install skills for your AI tool
specsafe install claude-code    # Tier 1: full skills
specsafe install aider          # Tier 2: conventions
specsafe install continue       # Tier 3: prompts

# Check project health
specsafe doctor

Skills Reference

21 canonical skills:

Planning Skills

Skill	Description
specsafe-brainstorm	Divergent exploration before converging on a solution
specsafe-principles	Product principles, non-goals, and quality priorities
specsafe-brief	Create a concise product brief — problem, users, scope, success criteria
specsafe-prd	Expand brief into full PRD with user journeys, FRs, NFRs
specsafe-ux	UX design foundations — tokens, components, accessibility, flows
specsafe-architecture	System architecture — components, data model, ADRs, technology stack
specsafe-readiness	Pre-development coherence and readiness check

Development Skills

Skill	Description
specsafe-new	Create a new spec with unique ID and initial requirements
specsafe-spec	Refine spec with detailed requirements, criteria, scenarios
specsafe-test	Generate test files from spec scenarios (all tests start skipped)
specsafe-code	TDD implementation — unskip tests one at a time, write code to pass
specsafe-verify	Run tests, validate against spec, loop on failure
specsafe-qa	Full QA validation with report and GO/NO-GO recommendation
specsafe-complete	Human approval gate — moves spec to COMPLETE

Utility Skills

Skill	Description
specsafe-init	Initialize project — creates dirs, config, PROJECT_STATE.md
specsafe-explore	Pre-spec exploration and research mode
specsafe-context	Gather and present project context for AI agents
specsafe-status	Project dashboard with spec counts by stage
specsafe-archive	Archive obsolete or abandoned specs
specsafe-doctor	Validate project health and configuration
specsafe-skill-creator	Create new SpecSafe skills with proper structure and validation

Agent Personas

Persona	Role	Archetype	Stages
Elena	Exploration Lead	Scout	EXPLORE
Kai	Spec Architect	Mason	SPEC, BRIEF, PRD
Reva	Test Engineer	Forge	TEST
Zane	Implementation Engineer	Bolt	CODE
Lyra	QA Inspector	Warden	QA, VERIFY
Cass	Release Manager	Herald	COMPLETE, STATUS, ARCHIVE, INIT, DOCTOR
Aria	UX Designer	Prism	UX
Nolan	System Architect	Sage	ARCHITECTURE

Supported Tools

Tool	Tier	Output
claude-code	1 — Full Skills	.claude/skills/*/SKILL.md + CLAUDE.md
opencode	1 — Full Skills	.opencode/skills/*/SKILL.md + .opencode/command/*.md
cursor	1 — Full Skills	.cursor/skills/*/SKILL.md + .cursorrules
continue	3 — Prompts	.continue/prompts/*.md + agent config
aider	2 — Conventions	.aider.conf.yml + CONVENTIONS.md
zed	2 — Conventions	.rules + .zed/settings.json
gemini	1 — Full Skills	.gemini/skills/*/SKILL.md + .gemini/commands/*.toml + GEMINI.md
antigravity	1 — Full Skills	.agent/skills/*/SKILL.md + .agent/rules/specsafe.md + AGENTS.md

Configuration

specsafe.config.json

{
  "project": "my-project",
  "version": "1.0.0",
  "tools": ["claude-code"],
  "testFramework": "vitest",
  "testCommand": "pnpm test",
  "coverageCommand": "pnpm test --coverage",
  "language": "typescript",
  "specsafeVersion": "2.0.0"
}

PROJECT_STATE.md

Tracks the current state of all specs in the project:

# Project State — my-project

## Active Specs
| ID | Title | Stage | Owner |
|----|-------|-------|-------|

## Completed Specs
| ID | Title | Completed |
|----|-------|-----------|

## Metrics
- Total specs: 0
- Active: 0
- Completed: 0

Further Reading

• Canonical Workflow — source of truth for the two-phase model
• Tool Support Guide — per-tool setup details
• Persona Guide — how each persona works