Agent-Native Demo¶
The main ForgeOS demo should show a product experience built around external code agents.
The user does not type a prompt into the demo app and wait for a hidden in-browser agent. Instead, the user creates or opens a ForgeOS app, chooses an external coding agent, and watches ForgeOS provide the context and evidence that make that agent effective.
Demo Story¶
- A user opens a ForgeOS app.
- The workroom shows the app state from
forge status,forge changed, andforge dev --once. - The user chooses Codex, Claude Code, or Cursor.
- ForgeOS shows the exact setup command and adapter state.
- The external agent edits the project in its own tool.
- Hooks and MCP feed Agent Memory.
- DeltaDB and Semantic Timeline connect file edits, commands, checks, and proofs.
- The workroom shows whether the change is ready for handoff or commit.
What The Demo Must Not Do¶
- It must not pretend the browser runs Codex, Claude Code, or Cursor.
- It must not present ForgeOS as a feature tour.
- It must not make DeltaDB, Agent Memory, or MCP the headline.
- It must not create fake internal tasks for Codex or Claude Code.
Those systems are the evidence layer. The product story is: external agents can work on this app safely because ForgeOS gives them context and verification.
Workroom Layout¶
| Area | Purpose |
|---|---|
| Project rail | App, branch, changed files, generated freshness, safe-to-edit state |
| Agent lane | Read-only transcript of the external agent session and launch/setup commands |
| App preview | Local preview of the app being changed, similar to Chef/Lovable, but without pretending the browser is the coding agent |
| Source focus | Authored files, generated files, review order, and current diff focus |
| Evidence stream | Hook events, MCP reads, timeline events, checks, proofs, handoff |
| Verification gate | generate --check, check, impact tests, verify --standard, handoff readiness |
The app preview is important. It gives the user the same fast visual feedback they expect from modern AI app builders. The difference is where the intelligence lives: Codex, Claude Code, or Cursor edits the project externally, while ForgeOS observes, verifies, and explains the work.
Demo Assets¶
The reproducible browser recording lives in marketing/demo/playwright-demo.html.
Regenerate assets with:
powershell -ExecutionPolicy Bypass -File marketing/demo/record-playwright-demo.ps1
The page is intentionally static and deterministic so release posts can be reproduced. It uses real ForgeOS command names and a curated scenario, but it does not claim to execute an external agent from the browser.