DARE Method¶
Design. Architect. Review. Execute. Methodology + CLI for AI-assisted software development, with mandatory human checkpoints.
DARE separates strategy (human) from tactics (AI) with explicit checkpoints: the human defines what and why and approves the plan; the AI implements how, iterating until tests/lint/types pass (the Ralph Loop).
flowchart LR
D[1. Design<br/>human defines] --> A[2. Architect<br/>AI proposes]
A --> R[3. Review<br/>human approves]
R --> E[4. Execute<br/>AI + Ralph Loop]
D -.-> dd[DESIGN.md]
A -.-> bb[BLUEPRINT.md]
R -.-> ap[✓ approval]
E -.-> cc[Code + green tests]| Phase | What | Who | Output |
|---|---|---|---|
| Design | the problem and the success criteria | human (AI assists) | DARE/DESIGN.md |
| Architect | architecture, contracts and tasks | AI proposes, human validates | DARE/BLUEPRINT.md |
| Review | explicit approval before spending tokens | human | ✓ approval |
| Execute | task-by-task implementation with the Ralph Loop | AI | code + green tests |
Start here¶
- Getting Started — install the CLI and run
dare init. - Greenfield — new project: design → blueprint → execute.
- Brownfield — legacy project: discover, reverse, dna, patterns, migrate.
- Configuration — the entire
dare.config.json. - CLI Reference — every command and flag.
- Knowledge Graph — the project's knowledge graph.
Quick install¶
npm install -g @dewtech/dare-cli
dare init meu-projeto
cd meu-projeto
dare design "Quero uma API de autenticação JWT"
What's new¶
- v3.8.0 — Formal Verification Gate: opt-in gate that PROVES (not just tests) marked critical modules against a Dafny spec (anti-bypass, exit 5).
- v3.7.0 — Brownfield Discovery: deterministic auto-discovery of patterns (
dare patterns) + lightweight planners. - v3.6.0 — Agent Hooks + Steering: event-driven automations + pattern injection via MCP.
- v3.5.0 — Dual Graph: Requirement↔Code graph +
dare graph owners/impact/trace/locate. - v3.4.0 — Security Hardening: hardened MCP server + publish with provenance.
- v3.3.0 — Reliable Verification Core: mutation testing, fail-to-pass, decay policy, best-of-N and
dare bench.
Details for each release in the CHANGELOG.