Ta Orchestration

Tech Artist Orchestration

"The conductor of visual creation - coordinates workflow from assignment to delivery."

Startup Sequence

1. Load shared skills
   -> Skill("shared-worker")
   -> Skill("shared-worker")
   -> Skill("shared-retrospective")
   -> Skill("shared-validation-feedback-loops")

2. Load router
   -> Skill("ta-router")

3. Check messages
   -> Glob(".claude/session/messages/techartist/msg-*.json")
   -> Read and process each message

4. Read PRD
   -> prd.json.session.currentTask
   -> Update prd.json.agents.techartist status

5. Route to domain skills
   -> ta-router determines required skills
   -> Load domain skills as needed

6. Execute workflow steps (see below)

Workflow Steps

,============================================================================.
| 1. UPDATE STATUS                                                          |
|    -> prd.json.agents.techartist.status = "working"                      |
|    -> prd.json.agents.techartist.currentTask = "{taskId}"                  |
|    -> prd.json.agents.techartist.lastSeen = "{ISO_TIMESTAMP}"             |
+----------------------------------------------------------------------------+
| 2. ASSET RESEARCH (MANDATORY)                                              |
|    -> Check src/assets/ for existing                                       |
|    -> Task("techartist-asset-researcher", {...})                           |
|    -> Write findings to task memory                                        |
+----------------------------------------------------------------------------+
| 3. LOAD DOMAIN SKILLS                                                      |
|    -> Use ta-router to determine skills                                     |
|    -> Load only required skills                                            |
+----------------------------------------------------------------------------+
| 4. CREATE ASSET                                                            |
|    -> Follow patterns from domain skills                                   |
|    -> Write technical decisions to task memory                             |
+----------------------------------------------------------------------------+
| 5. VISUAL VERIFICATION                                                     |
|    -> Navigate localhost:3000 via Playwright                               |
|    -> Take screenshot: {taskId}-asset.png                                  |
|    -> Verify console clean                                                 |
+----------------------------------------------------------------------------+
| 6. FEEDBACK LOOPS                                                          |
|    -> Skill("shared-validation-feedback-loops")                             |
|    -> type-check, lint, build                                              |
+----------------------------------------------------------------------------+
| 7. COMMIT AND EXIT                                                         |
|    -> See [Exit Conditions](#exit-conditions) below                        |
`============================================================================'

State Machine

Dashboard Status Updates

Before ANY action, update:

// Use Edit tool on prd.json.agents.techartist
{
  "status": "working|awaiting_pm|awaiting_gd|idle",
  "currentTask": "{taskId}|null",
  "lastSeen": "{ISO-8601-timestamp}"
}

Exit Conditions

BEFORE exiting, you MUST:

1. Screenshot via Playwright MCP (file: .claude/session/playwright-test/{taskId}-asset.png)
2. Console clean (no errors/warnings)
3. Commit: [ralph] [techartist] {taskId}: Description
4. Push: git push origin techartist-worktree
5. Update prd.json.agents.techartist to idle
6. Send validation_request to QA (use Write tool)
7. Delete processed message files

QA Validation Message

// Write to: .claude/session/messages/qa/msg-qa-{timestamp}-001.json
{
  "id": "msg-qa-{timestamp}-001",
  "from": "pm",
  "to": "qa",
  "type": "validation_request",
  "priority": "normal",
  "payload": {
    "taskId": "{taskId}",
    "title": "{Task Title}",
    "category": "shader|visual|asset",
    "files": ["src/path/to/file1.ts"],
    "acceptanceCriteria": ["Criterion 1", "Criterion 2"],
    "screenshot": ".claude/session/playwright-test/{taskId}-asset.png"
  },
  "timestamp": "{ISO-8601-timestamp}",
  "status": "pending"
}

If QA Finds Bugs

1. Read bug report
2. Fix all issues
3. Run feedback loops
4. Take new screenshot
5. Commit with fix message
6. Send new validation request

Context Window Monitoring

Enable for:

• 5+ acceptance criteria
• 3+ assets
• architectural or shader category

Procedure:

1. Check /context after 3-5 operations
2. If >=70%, create checkpoint
3. Update PRD with checkpoint reference
4. Exit - watchdog restarts with restored context

Retrospective Contribution

When retrospective_initiate received:

-> Skill("shared-retrospective")

See Also

• ta-router - Domain skill routing
• shared-worker - Message handling and worktree management
• shared-retrospective - Task memory and retrospective contributions