ForgeOne Flow

See how your AI workspace is actually wired — agents, prompts, skills, and tools on one interactive map.

Copilot agents, prompt files, skills, MCP tools, and instruction files pile up fast across .github/, .vscode/, and .cursor/. Nobody can tell which agent has which tools, which prompt is orphaned, or what's missing. ForgeOne Flow reads those files and draws the map, then tells you what to fix.

Why Flow

• One picture of your AI setup. Agents, prompts, skills, tools, and instruction files with the edges between them, plus warnings for broken or missing links.
• Advice you can apply. Flow suggests the fix (for example, a missing paired prompt) and creates it without leaving the map.
• Proof for pull requests. Copy an Advice digest into a PR so reviewers see what changed in the AI workspace.
• Zero setup to start. No init, no config: open the map on any repo.
• Optional governance. If your team wants it, track stages, human gates, and evidence in .forgeone/, and let agents report through MCP.

Quick start

1. Install ForgeOne Flow from the Extensions view, or run ext install anghelll.forgeone-flow.
2. Open the ForgeOne Flow view in the Activity Bar.
3. Under AI workspace, click Open Workspace Map (or run ForgeOne Flow: Open Workspace Map).

Want a populated example? Open examples/workspace from this repo.

The map is an inventory of what's on disk. It does not trace live chat sessions.

Features

Workspace Map and Advice

Primary target is VS Code + GitHub Copilot (agents, prompt files, skills, MCP tools, copilot-instructions.md). Cursor rules, AGENTS.md, and CLAUDE.md are detected too.

Advice actions such as Create paired prompt and Open agent (tools) close the gap in place. The ideal overlay shows the expected result as a ghost before you apply it:

Copy digest turns the current advice into Markdown for a PR description.

Multi-agent Stage

The Orchestration reading draws how your custom agents are declared to work together: the orchestrator, the subagents it may call (agents:), and handoffs with their label, prompt, send, and model. The inspector shows each agent's model fallback list, tools grouped by risk, invocation flags, and which repo instruction surfaces it always receives.

• Walkthrough (W) steps through declared edges only, in a fixed order, and exports as one SVG per step plus walkthrough.md.
• Present Multi-agent Stage opens the map in a projector-friendly theme in its own window. Rehearsal checklist: docs/PRESENTATION.md.
• What Flow does not do: run agents, read chat sessions, or report live status. Every edge comes from a file and field you can open. A handoff to an agent that doesn't exist is shown as not found. Edges guessed from next: or names stay hidden until you turn them on (P).

Settings: forgeoneFlow.map.renderer (app or legacy) and forgeoneFlow.map.theme (stage or editor). The map UI uses React Flow and ELK.

Scaffolding

• Scaffold Workspace Pack adds a map-ready set of agents, prompts, and skills, with a dry-run preview before anything is written.
• Scaffold Agent Template starts a single agent; Export Copilot/Cursor Agent exports one for either tool.

Governance (optional)

For teams that want an explicit pipeline around AI-assisted work:

• Init Workflow creates .forgeone/ with stages and pending gates.
• Record Gate marks a human approval; Verify-record logs which checks ran and passed.
• Pipeline view shows the timeline of gates and evidence.
• Context trail records which context packs and instruction files the agent used.
• Ecosystem shows workflow health across several repos.
• Hooks (opt-in) warn on commit when gates are pending. They never block.
• MCP server lets the agent read status and record context, verification, and gates (by: agent:<name>). Run ForgeOne Flow: Copy MCP Config to get the configuration for .vscode/mcp.json or .cursor/mcp.json.

Nothing in .forgeone/ is required to use the map.

Install

• VS Code: Visual Studio Marketplace.
• Cursor and other Open VSX editors: Open VSX — search ForgeOne Flow in Extensions.

UI language follows your editor; override it with forgeoneFlow.uiLocale (auto, en, or es).

Part of ForgeOne

ForgeOne Flow comes from the same publisher as the ForgeOne BDD extensions. Flow maps your AI workspace; the BDD extensions work on .feature files and tests:

• BDD Guardian: step ↔ binding navigation and diagnostics
• BDD Pilot: run Reqnroll / SpecFlow scenarios
• BDD Gherkin Format: format and highlight .feature files

CLI

The same features are available from a clone of this repo. The npm package is not published yet.

npm install
npm run forgeone -- workspace map
npm run forgeone -- workspace advice --markdown   # same as Copy digest
npm run forgeone -- workspace lens --file src/app.ts   # which instructions apply here, per platform, and why
npm run forgeone -- workspace agents --redact         # declared agents, handoffs, subagents; names and paths redacted

.forgeone/
  workflow-state.json    # stage and pending gates
  events.jsonl           # append-only history
  workflow-def.json      # custom stages and gates
  context-pack.json      # context-pack-v1 schema

Learn more

• Demo walkthrough
• Presenting multi-agent mode
• Adoption guide
• What's new
• Product definition and Roadmap
• Release process

Found a bug or have an idea? Open an issue. MIT License.