Skip to content

ARCHITECTURE.en.md

Latest commit

 

History

History
42 lines (33 loc) · 1.58 KB

ARCHITECTURE.en.md

File metadata and controls

42 lines (33 loc) · 1.58 KB

Architecture

End-to-End Flow

flowchart LR
  A["Requirement Input"] --> B["Brainstorm (multi-model)"]
  B --> C["Debate (N rounds)"]
  C --> D["Synthesis (selected model)"]
  D --> E["Roadmap (Round -> Task)"]
  E --> F["Task Compiler (TaskSpec/Checks/L2 Skill/MCP allowlist)"]
  F --> G["Execution Worker"]
  G --> H["Verification (tests/lint/build)"]
  H --> I["Reviewer Acceptance"]
  I --> J["Pass -> Next Task"]
  I --> K["Fail -> Rework"]
Loading

Relationship Table

Stage Owner Role Inputs Outputs MCP/Tooling
Brainstorm Brainstormers (multi-model) Requirement candidate proposals model-router
Debate Brainstormers proposals critiques per round model-router
Synthesis Planner model proposals + critiques architecture decision + roadmap model-router
Task Compile Brain/Orchestrator roadmap node TaskSpec.md, Checks.md, TASK.SKILL.md, mcp_allowlist.json scripts/brain.ts
Execution Worker model task package code changes + explanation claude-runner or provider API
Verification CI/local code changes pass/fail evidence test/lint/build commands
Acceptance Reviewer model diff + evidence accept/reject + risks provider API / MCP

Skill Layers

  • L0: roadmap scheduler only
  • L1: domain reusable playbooks
  • L2: short task instance skill generated per task

Permission Model

  • Default: minimum required MCP tool allowlist per task
  • Optional: dangerous permission switches by model (codex/claude/gemini)
  • Full automation mode: dangerous switches all enabled