DecoupleFlow — Design skill for Claude Code
DecoupleFlow brings lightweight soft orchestration to your existing agents.
How to install DecoupleFlow
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Silicontoy/DecoupleFlow and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What DecoupleFlow does
DecoupleFlow brings **lightweight soft orchestration** to your existing agents. Its independent **visual state-machine editor** turns states, transition conditions, and quality gates into readable Markdown and portable Skills for Codex, Claude Code, or your own harness.
Alternatives in Design
- Frontend Design — Create distinctive, production-grade frontend interfaces with high design quality 94.1k ★
- Remix — Restyle a site in another design vocabulary — brutalist, swiss, art-deco, cyberpunk, soft-ui, or editorial 3.6k ★
- Drawbridge — Design editor for Claude Code and Cursor 960 ★
README
DecoupleFlow
**Quick Build Long Run**
DecoupleFlow brings **lightweight soft orchestration** to your existing agents. Its independent **visual state-machine editor** turns states, transition conditions, and quality gates into readable Markdown and portable Skills for Codex, Claude Code, or your own harness.
You define what needs to happen and when the process can move forward. Your agent decides how to do the work, chooses its tools, and carries context between steps. The Skill validates the agent's transition decisions and saves progress so long-running tasks can continue across sessions.
- Guide the process: define states, shared checklist conditions, branches, loops, and waits while preserving the agent's choice of execution method.
- Take it with you: Markdown + Mermaid, Agent Skills, LangGraph, or Dify exports.
- Pick up where you stopped: file-backed instances, evidence checks, and a live execution view.
Like giving an AI a senior teammate's working notes: a clear process, with room to decide how to carry it out.
[Quick start](#quick-start) · [Try three demos](#try-three-demos) · [Codex close/resume example](examples/codex-resume/) · [Tested integrations](examples/verified/) · [MIT](LICENSE) · [简体中文](README.zh-CN.md)
*A 45-second recording of the real interface and exported Skill runtime, using example task data. [Still screenshots and capture details](docs/media.md).*
Try three demos
Choose one of the first three cards in **New machine**. Each includes sample inputs, a ready-to-import definition, and Skills for Codex, Claude Code, and Copilot.
| Demo | Start with → finish with | Try it |
|---|---|---|
| Fix a bug, prove it | A cart total of 17 instead of 42 → a focused patch and three passing regressions | [Guide](examples/demos/ |
Related Skills
Factory Init
Interactively design and scaffold an autonomous agent loop in this project (goal, gate, state, stop conditions
Yara Code
Yara Code — a lightweight code editor for agent-driven development: a terminal UI and a GPU window that mirror
Design Loop
Run the full design iteration loop (harness component 06): planner → generator → four machine gates → critic,
Deep Plan
Builds one feature end to end in a project that already exists: reads codebase, GRILLS user, UI mock sign-off,
Design Critique
UX heuristic review of Figma screens — evaluates against Nielsen's 10 heuristics, visual design quality, spaci
Gsd T Design Review
GSD-T: Design Review — Independent Review Agent (Term 2)
Related Agents
Docs Human
Human-readable documentation writer. Maintains DECISIONS.md, STATUS.md, and ARCHITECTURE.md at the repo root —
Intraday PM
tradefloor intraday portfolio manager. Makes the final intraday calls after the risk committee, writes the pic
Analyst Writer
Analyst-writer that turns technical material - research results, task breakdowns, incident post-mortems, archi
