# Current Process - Agentic PM Demo

This side document records the process currently being used to finish the Agentic PM Demo UX and automation pass on branch `codexMacTiger/agentic-pm-next`.

## Working Objective

We are completing a 10-point UX and behavior goal:

1. Move model settings into a popup that can be hidden behind a button after setup.
2. Run a model connection test when the API key/config is provided.
3. Highlight each work package while running, then mark completed packages with a green done/version indicator.
4. Make the work-package detail view output-first, with markdown rendering.
5. Hide non-output package details inside dropdown/collapsible sections.
6. Show detailed model request and response traces in the agent log.
7. Make the chat area independently scrollable, like the Codex desktop chat.
8. Show current running activity and hints inside the chat area.
9. Update the page title and chat title when the user provides a product name.
10. Run a product-idea test and review the generated outputs for quality.

## Current Workflow

The working loop is:

1. Inspect current source and runtime state before trusting previous context.
2. Make narrowly scoped UI or behavior changes in the existing Next.js app.
3. Validate with automated commands:
   - `npm test`
   - `npm run lint`
   - `npm run build`
4. Validate runtime behavior in the browser at `http://127.0.0.1:3000`.
5. Use the product idea below as the real end-to-end scenario:
   - `Build a product called TorqueGuard AI for industrial tightening anomaly monitoring with traceable shift dashboards, maintenance diagnostics, and operator alerts.`
6. Review generated package output quality by opening work-package detail views and checking that outputs are concrete, package-specific, markdown-rendered, and traceable to the product idea.

## Current Evidence

Observed browser/runtime evidence:

- Model settings appear as a popup panel behind `Live Settings`.
- The connection state shows `Connected to the configured model.`
- The active product title is `TorqueGuard AI` in the app header and chat header.
- Work packages reached `Done` state and show `v1` version badges.
- The agent log shows board automation completion and package/detail activity.
- Customer Requirements Specification detail view shows `PRIMARY OUTPUT` first.
- The CRS output is markdown-rendered into headings and tables.
- Package context is hidden behind a `PACKAGE CONTEXT` collapsible section.
- CRS output includes concrete TorqueGuard AI requirements for operators, supervisors, maintenance engineers, plant managers, legal/compliance, security, technology, manufacturing, logistics, and trade.

Fresh command evidence from the last verification pass:

- `npm test`: 6 test files passed, 15 tests passed.
- `npm run lint`: completed with exit code 0.
- `npm run build`: completed successfully with Next.js production build output.

## Output Review Notes

The full board automation completed and generated outputs for:

- CRS
- Final Concept
- SRS
- Safety
- Certification
- Test
- Decompose
- Industrial Design
- Patent
- Design FMEA
- Engineering Concept
- Service Review

Quality review status:

- The CRS output is satisfactory for an MVP demo: it is concrete, product-specific, structured, and rendered as tables/headings.
- The board-level completion log confirms all 12 package outputs were generated.
- Further detailed review can sample each package detail view individually if a stricter artifact-by-artifact approval gate is needed before remote deployment.

## Important Constraints

- Do not hardcode or commit the user API key.
- Keep the app mock-capable when no API key is provided.
- Treat `execute` as generated by the configured model, not as a real engineering/certification/toolchain execution.
- Preserve the current Codex-like visual language: quiet, dense, soft panels, small controls, no heavy board lines.
- Keep the current local dev target at `http://127.0.0.1:3000`.

## Next Step

Run one final verification pass after this document is added, then report the current state clearly. If the user wants stricter output review, open each work-package detail view and record a per-package approval note.