Designs, audits, and improves the AgentDoctor dashboard as a serious developer product intelligence UI—not a raw JSON/API debug console. Use when working on the AgentDoctor dashboard, dashboard UI/UX, navigation, project overview, Project Brain/Intelligence UI, security/architecture/AI-agent/student/evidence/proof UI, developer workflow surfaces, or dashboard information architecture.
Installs into .claude/skills of the current project.
Are you the author of Agentdoctor Dashboard Product Design?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/pranee54-agentdoctor-dashboard-product-design)
---
name: agentdoctor-dashboard-product-design
description: >-
Designs, audits, and improves the AgentDoctor dashboard as a serious developer
product intelligence UI—not a raw JSON/API debug console. Use when working on
the AgentDoctor dashboard, dashboard UI/UX, navigation, project overview, Project
Brain/Intelligence UI, security/architecture/AI-agent/student/evidence/proof UI,
developer workflow surfaces, or dashboard information architecture.
---
# AgentDoctor Dashboard Product Design
AgentDoctor is a software project intelligence and assurance platform.
The dashboard must help a user understand their project, find problems, learn how it works, inspect evidence, safely make changes, and verify those changes.
The dashboard must NOT look like:
- a raw JSON API viewer
- an admin CRUD panel
- a developer debug console
- a generic SaaS template
- a monitoring dashboard full of meaningless charts
It should feel like a serious engineering product.
**Full design bible:** [reference.md](reference.md) — follow it for Home, navigation, components, truth model, security UX, student UX, visual language, audit matrix, and verification.
## Absolute rules
1. **Do not invent backend capabilities.**
2. Before changing UI, inspect existing API routes, contracts, dashboard handlers, types, and frontend implementation.
3. **Backend truth is authoritative.**
4. Do not fabricate scores, metrics, evidence, or pages.
5. Preserve safety: approvals, path safety, workspace isolation, truth labels, evidence. No unrestricted shell.
6. Keep API contracts backward compatible.
## Canonical code locations
Inspect first (paths may grow; start here):
| Area | Location |
| ---- | -------- |
| Embedded dashboard | `src/dashboard/server.ts` |
| Package version | `src/constants.ts` → `PACKAGE_VERSION` |
| Status / ops health | `src/ops/health.ts`, `/api/status` |
| Product APIs | `src/product/**`, dashboard `/api/*` handlers in `server.ts` |
| Chat | `/api/chat` + `src/agent/chat/**` |
| Tests | `tests/unit/product/dashboard-*.test.ts`, e2e if present |
## Mandatory workflow
Copy and track:
```
Dashboard task:
- [ ] 1. Read this skill + reference.md sections relevant to the change
- [ ] 2. Inventory routes/endpoints/types from code (not memory)
- [ ] 3. Sample real API responses (nullable/optional fields)
- [ ] 4. Audit current page(s) against the design hierarchy
- [ ] 5. Implement UI consuming real contracts only
- [ ] 6. Loading / empty / error / technical-details states
- [ ] 7. Version labels from PACKAGE_VERSION / status API only
- [ ] 8. Verify (suite + browser + no overflow / broken routes)
- [ ] 9. Final output block (see below)
```
### Design hierarchy (always)
```
PROJECT → UNDERSTAND → CHECK → ACT → VERIFY
```
User must answer in ≤30 seconds:
1. What project am I looking at?
2. What is the current state?
3. Is anything important wrong?
4. What should I do next?
5. Where can I investigate deeper?
Progressive disclosure: simple first → technical second → raw JSON only in expandable Technical Details.
### Version consistency (concrete defect)
Never hardcode `AgentDoctor 2.0` when package is `3.0.0`.
One authoritative source: `PACKAGE_VERSION` / status API `ops.version` (or equivalent). Audit title, header, footer, about, metadata.
### Raw JSON
Never primary Home. Use structured overview cards with semantic states (`Healthy` / `Attention` / `Warning` / `Blocked` / `Unknown` / `Not analyzed`) — **not invented percentages**.
### Navigation
Group conceptually (OVERVIEW / UNDERSTAND / CHECK / INTELLIGENCE / OPERATIONS / LEARNING / SYSTEM). **Only expose routes that exist.** Label experimental honestly. No fake pages.
### Truth labels
Preserve `VERIFIED` | `INFERRED` | `UNKNOWN` | `EXTERNAL`. Never imply certainty the backend does not have.
## Implementation preferences
- Prefer improving the existing embedded dashboard (`src/dashboard/server.ts` HTML/CSS/JS) unless the repo already has a separate SPA framework.
- Do not introduce a large new UI framework without a clear technical reason.
- Reuse patterns: Layout, Sidebar, StatusBadge, cards, EmptyState, TechnicalDetails, EvidencePanel.
- Dark professional engineering UI: strong hierarchy, restrained accents, no template aesthetics.
## Verification
After changes:
1. `npm run verify` (or at least typecheck + relevant dashboard tests)
2. Browser: Home + every nav item
3. Empty / no-git / no-tests / no-AI / with-AI / security findings when practical
4. Technical details expand + copy
5. Responsive: no horizontal overflow, no clipping
6. All version labels consistent
7. No console errors; no broken routes
Do not claim "production ready" unless verification supports it.
## Final output (required)
```markdown
1. Dashboard audit summary
2. Files changed
3. Routes audited
4. UX problems fixed
5. Backend contracts preserved
6. Tests run
7. Remaining limitations
8. Screenshots/browser verification performed
```
## Goal
Turn AgentDoctor's existing engineering capabilities into a clear, trustworthy, professional project intelligence interface — not merely prettier chrome.