OpenFlow: The Governance Layer for AI-Driven Development

中文文档 · Full handbook

OpenFlow is a documentation-governed workflow for AI-assisted development on OpenCode. It does not ask the AI to start with “how should we code this?” It first asks:

• What is the exact boundary of the change?
• Which existing constraints must not move?
• What evidence proves the work is actually done?
• Where will the decision live after the chat session disappears?

It is designed for brownfield systems where traceability, verification evidence, and long-term project memory matter more than the fastest first draft.

Why OpenFlow exists

AI agents can write code quickly, but unmanaged AI-driven changes often create three problems:

1. Scope drift — the agent fixes one thing and quietly rewrites another.
2. Unverifiable completion — “done” becomes a chat message instead of fresh lint/typecheck/test evidence.
3. Lost rationale — months later nobody knows why a file, function, or behavior exists.

OpenFlow turns documentation into an executable constraint system:

• docs/current/ stores current system facts the AI must obey.
• docs/changes/ stores active feature workspaces and change boundaries.
• docs/decisions/ stores cross-feature architectural decisions.
• docs/archive/ stores immutable completed history and requirement-to-code traceability.

Core workflow

brainstorm → feature → writing-plan → implement → quality-gate → archive

| Stage | What happens | User entrypoint | |---|---|---| | Brainstorm | Explore requirements, options, and trade-offs before formalizing work. | Ask the agent to brainstorm, or invoke the openflow-brainstorm skill. | | Feature | Clarify the change boundary and generate design/behavior documents. | /openflow-feature <description> | | Writing plan | Convert design docs into a structured implementation plan. | /openflow-writing-plan <feature> | | Implement | Create an implementation run and delegate execution. | /openflow-implement <feature> | | Quality gate | Verify evidence, risk, freshness, and readiness. | AI calls openflow-quality-gate automatically. | | Archive | Freeze history, promote current facts, and generate traceability. | /openflow-archive <feature> |

What you get

• Requirement-to-code traceability — completed features produce an implementation-mapper.md mapping requirements to changed files, functions, and symbols.
• Evidence-led readiness — completion requires fresh verification, not a verbal claim.
• Long-term project memory — project facts survive agent switches, developer handoffs, and lost chat history.
• Safer AI execution — current docs, decisions, and behavior constraints are injected back into implementation.
• Optional orchestration — works with OpenCode by default and can integrate with oh-my-openagent (omo) and GitNexus for deeper multi-agent/code-intelligence workflows.

Quick install

For LLM agents such as Claude Code, Cursor, Trae, or OpenCode, paste this instruction into the agent:

Install and configure OpenFlow by following the instructions here:
https://fastknifes.github.io/openflow/getting-started/installation

Manual install:

npm install @fastknife/openflow

Then enable the plugin in ~/.config/opencode/opencode.json or opencode.jsonc:

{
  "plugin": ["@fastknife/openflow"]
}

Initialize your project in OpenCode:

/openflow-init

Then start your first governed feature:

/openflow-feature add user profile page

Learn how to use it

• 10-minute quickstart
• Installation guide for AI agents
• Command reference
• Feature workflow guide

License

MIT License. Developed by fastknife.