Orchestrator Execution

Execute planned tasks, manage workflow status, and update the knowledge graph with what was learned.

Workflow

•Fetch task with sections and graph context (single call)
•Dispatch to recommended agent with task context + graph context
•Wait for completion (never interrupt)
•Run verification commands
•If failed: keep IN_PROGRESS, surface issue, stop here
•If passed: update the knowledge graph
•Advance status
•Check siblings, suggest next task

MCP Tools

Fetch task with sections and graph context:

code

query_container
  operation: "get"
  containerType: "task"
  id: <task-uuid>
  includeSections: true
  includeGraphContext: true

Query workflow state (REQUIRED before status change):

code

query_workflow_state
  containerType: "task"
  id: <task-uuid>

Advance status:

code

advance
  containerType: "task"
  id: <task-uuid>
  version: <version-from-workflow-state>

Get next task:

code

get_next_task
  featureId: <feature-uuid>

Context to Pass to Sub-Agents

Include these sections verbatim:

•Implementation Plan
•Context Files
•Verification
•Dependencies and Constraints

Also include the graphContext from the query response. Format it as a readable summary at the top of the agent prompt:

code

Knowledge graph context for this task:

Molecule: <name> — <knowledge>
  Atom: <name> — <knowledge>
    matched: file-a.ts, file-b.ts
    related: <other atom> — <reason>
Orphan atoms: <atoms matched but ungrouped>
Unmatched files: <files in Context Files no atom covers>

Status Rules

Never skip statuses. Always:

•Call query_workflow_state first
•Only use status from allowedTransitions
•Pass version field

Typical flow: NEW -> ACTIVE -> CLOSED (configurable via config.yaml pipelines)

Agent Defaults

Work Type	Agent / Model
Simple implementation	sonnet
Complex implementation	opus
Test writing	sonnet
Exploration	Explore / haiku
Code review	code-reviewer / sonnet

Graph Update After Verification (REQUIRED)

After verification passes and before advancing status, update the knowledge graph. This is not optional — the skill enforces it.

Step 1 — Identify affected atoms

Check which atoms match the files touched during the task. Use query_graph context if not already loaded, or reference the graphContext from step 1.

Step 2 — Update atom knowledge

If the work changed how the area works (new patterns, new constraints, changed integration points):

code

manage_graph
  operation: "update"
  entityType: "atom"
  id: <atom-uuid>
  knowledgeMode: "append" | "overwrite"
  knowledge: "<what changed about this area>"
  lastTaskId: <task-uuid>
  version: <current-version>

Use append for incremental additions. Use overwrite after major refactors.

Step 3 — Update atom paths

If files were moved, new directories created, or the atom's boundary changed:

code

manage_graph
  operation: "update"
  entityType: "atom"
  id: <atom-uuid>
  paths: '["src/new/path/**"]'
  lastTaskId: <task-uuid>
  version: <current-version>

Step 4 — Create new atoms for uncovered areas

If the task created files in an area no atom covers (visible from unmatchedPaths):

code

manage_graph
  operation: "create"
  entityType: "atom"
  projectId: <project-uuid>
  moleculeId: <molecule-uuid if applicable>
  name: "<area name>"
  paths: '["src/new/area/**"]'
  knowledge: "<how this area works, patterns, constraints>"
  createdByTaskId: <task-uuid>

Step 5 — Update related atoms

If the work revealed or changed connections between atoms:

code

manage_graph
  operation: "update"
  entityType: "atom"
  id: <atom-uuid>
  relatedAtoms: '[{ "atomId": "...", "reason": "consumes events from" }]'
  lastTaskId: <task-uuid>
  version: <current-version>

Step 6 — Append changelog entries for significant changes

code

manage_changelog
  operation: "append"
  parentType: "atom"
  parentId: <atom-uuid>
  taskId: <task-uuid>
  summary: "<what changed and why>"

Not every atom update needs a changelog entry. Use judgment:

•Minor knowledge tweaks: skip, the knowledgeMode=append separator traces it
•New atom created: skip, createdByTaskId traces it
•Meaningful behavior changes: log it
•Changed integration points between atoms: log it

Step 7 — Advance status

After graph updates are complete, proceed with status advancement:

code

query_workflow_state
  containerType: "task"
  id: <task-uuid>

code

advance
  containerType: "task"
  id: <task-uuid>
  version: <version>

After Task Completion

•Check if all sibling tasks complete
•If yes: prompt to advance feature status
•Suggest next task