# Multi-Agent Production Studio

Webdev Agent OS v11 coordinates specialist production roles through a shared, auditable work graph.

## Purpose

The studio prevents a common failure of multi-agent systems: several impressive outputs that do not combine into one reliable product. Each production task therefore carries an owner, dependencies, acceptance criteria, budget, evidence, conflicts, and approval rules.

## Specialist roster

The default roster includes strategy, UX research, visual design, content, frontend, backend, integration, QA, accessibility, security, SEO, analytics, and deployment agents. Each role has a dedicated charter under `prompts/agents/`.

The roster represents governed production responsibilities. The seed data does **not** claim that an external AI model executed any task. A run records its provider and model as `not-connected` until real execution evidence is supplied.

## Production contract

A studio project defines:

- Business objective and measurable outcome
- Intended audience
- Constraints and quality target
- Responsible human owner
- Current stage and status

A task defines:

- Project, discipline, and assigned specialist
- Dependencies and acceptance criteria
- Inputs, outputs, and evidence references
- Estimated and consumed time/cost
- Whether human approval is mandatory
- Current status and blocking reason

## Dependency engine

Tasks are refreshed into `ready` or `blocked` states from their dependency graph. Cyclic task graphs are rejected. An unresolved conflict can block an otherwise-ready task.

The standard production plan is:

1. Product strategy
2. UX journeys
3. Visual direction
4. Content system
5. Frontend vertical slice
6. Backend contract
7. Integration adapters
8. Accessibility verification
9. Security and privacy review
10. SEO baseline
11. Analytics and measurement
12. QA and regression verification
13. Deployment and smoke testing

These are dependencies, not a rigid waterfall. Independent tasks can proceed in parallel when their evidence contracts are clear.

## Evidence and approval

A task cannot be submitted without an evidence reference. Approval-required tasks move to `awaiting-approval`; a reviewer or owner must approve or request changes. The specialist that performs work cannot satisfy the reviewer gate merely by marking its own work complete.

The system distinguishes:

- Example evidence
- Operator-recorded evidence
- Connected model evidence
- Verified test or deployment evidence

## Conflict resolution

Cross-discipline disagreement is stored as a first-class conflict with:

- Affected agents and tasks
- Severity
- Available options
- Human resolution
- Rationale and decision history

This is intended for real tradeoffs such as visual richness versus performance, analytics detail versus privacy, speed versus maintainability, or SEO content versus product clarity.

## Budget governance

A production budget records maximum, committed, and consumed hours and cost. It is planning evidence, not a replacement for accounting. Reserved capacity and commercial billing remain governed by the Orchestration and Agency workspaces.

## Safety boundary

Agents may prepare work, evidence, and recommendations. They may not silently:

- Deploy to production
- Approve their own gated work
- Contact a client
- Commit spending
- Alter protected data
- Bypass security, privacy, authorization, or release gates
- Claim tests, measurements, or model execution without evidence

## API

- `GET /api/studio`
- `GET /api/studio/stats`
- `GET /api/studio/export`
- `POST /api/studio/projects`
- `POST /api/studio/projects/:id/plan`
- `POST /api/studio/tasks`
- `POST /api/studio/tasks/:id/start`
- `POST /api/studio/tasks/:id/submit`
- `POST /api/studio/tasks/:id/decision`
- `POST /api/studio/conflicts`
- `POST /api/studio/conflicts/:id/resolve`
- `POST /api/studio/budgets`
- `POST /api/studio/report`

## CLI

```bash
npm run studio:stats
npm run studio:report
node scripts/studio-cli.js view
node scripts/studio-cli.js plan <project-id>
```

## Storage

- `config/studio-store.json` — working store
- `config/studio-seed.json` — reset/example baseline
- `config/studio-store.schema.json` — structural contract
- `reports/multi-agent-studio/` — generated reports