| name | kanban-orchestrator |
| description | Kanban board creation and workflow decomposition persona. Trigger phrases: "orchestrator", "dispatch workers", "kanban workflow creator". |
Kanban Workflow Creator / Orchestrator — hermes-ops Deployment Guide
Runtime skill: ~/.hermes/skills/autonomous-ai-agents/kanban-orchestrator/SKILL.md
The Kanban Orchestrator is the routing layer between intent and execution. It:
- Takes a goal and decomposes it into structured kanban tasks
- Fans work items out to the appropriate worker lanes
- Verifies output and closes the loop
Trigger
"orchestrator" or "run the orchestrator"
"dispatch workers", "fan out", "workstream"
"kanban workflow creator" — creates board structure from a goal
Two Modes
Mode 1: Workflow Creator
Given a goal → decompose into a full kanban board with lanes.
User: "build a paper ingestion pipeline"
↓
Orchestrator creates:
- Backlog lane
- t_001: Parse PDF metadata
- t_002: Extract figures and tables
- t_003: Generate S2ORC JSON
- Doing lane
- Done lane
Mode 2: Orchestrator Dispatch
Takes output from Research Council or TWG → fans out to workers.
Research Council synthesizes: "3 work items identified"
↓
Orchestrator:
- hermes kanban create for each item
- Assigns to worker lanes (codex, orcaid, subagents)
- Runs verification loop
- Closes loop on completion
Deployment (hermes-ops)
Step 1 — Load the skill
The runtime skill lives in ~/.hermes/skills/autonomous-ai-agents/kanban-orchestrator/SKILL.md. No installation needed.
Step 2 — Activate
hermes profile chat --profile kanban-orchestrator
Step 3 — Provide a goal
The orchestrator reads the goal and:
- Decomposes into discrete work items
- Creates kanban tasks with
--body containing the task definition
- Assigns lane priority (backlog → ready → doing → done)
- Sets WIP limits per lane
- Dispatches initial items to worker lanes
Step 4 — Monitor and close
hermes kanban list # track progress
hermes kanban complete t_XXXX --result "..." # close with verification
hermes kanban heartbeat t_XXXX # signal alive
Orchestrator Dispatch Loop
WHILE unverified_items:
item = select_next(kanban)
worker = assign(item, worker_pool)
result = worker.execute(item)
IF verification_passed(result):
kanban.complete(item, result)
notify(user)
ELSE:
kanban.block(item, reason)
notify(orchestrator, "rework needed")
Profile Files
~/.hermes/profiles/kanban-orchestrator/SOUL.md — identity, dispatch loop, drift correction
~/.hermes/skills/autonomous-ai-agents/kanban-orchestrator/SKILL.md — runtime execution
See Also
- [[two-council-architecture]] — orchestrator stack position
- [[delegation-verification]] — verification loop pattern
- [[orchestrator-memory]] — verified outcomes and drift logs