humanerd-drew/opencode-drewgent/kanban-worker
Claude Code and agent skill by humanerd-drew/opencode-drewgent
- community
Checking that the SKILL.md file is still in the repository.
Install
npx vskill@latest install humanerd-drew/opencode-drewgent/kanban-workerSKILL.md
The same SKILL.md was published earlier as humanerd-drew/opencode-drewgent/kanban-worker.
title: Kanban Worker name: kanban-worker type: document space: concept tags: [concept] created: 2026-05-20 updated: 2026-06-10 links:
- "[[@action/skills/SKILL-INDEX]]"
- "[[@memory/growth/{{AGENT_NAME_LOWER}}-kanban-implementation-plan]]"
- "[[kanban-dashboard]]"
- "[[kanban-orchestrator]]"
- "[[references/protocol.md]]"
- "[[@identity/brain/rules]]"---
Kanban Worker Skill
Task execution skill for kanban worker agents. A worker agent picks up tasks from the {{AGENT_NAME}} kanban board, executes them, and reports results.
Overview
A kanban worker is an autonomous agent that:
- Claims a task from the board (or receives a task_id via env)
- Reports heartbeat while working
- Executes the task (code, research, writing, etc.)
- Reports completion with result/summary
Workers are dispatched by the kanban dispatcher cron job or receive tasks via KANBAN_TASK_ID in the environment.
Workflow
Startup — Acquire Task
A worker can acquire a task in two ways:
Option A: Dispatcher assigns task_id via env
Read KANBAN_TASK_ID to know which task to work on.
Option B: Self-claim from board
During Work — Heartbeat
Long-running tasks must send heartbeats every ~60 seconds:
Heartbeat purpose:
- Prevents claim expiry (default TTL: 1 hour)
- Lets dispatcher monitor progress
- Provides audit trail in task_events
Task Completion
When done, call kanban_complete:
Hallucination check: created_cards IDs are verified against the DB. Any fake IDs cause a completion_blocked_hallucination event and exception.
Task Failure
If task cannot be completed:
Consecutive failures are tracked. After max_retries failures, task is moved to failed state.
Blocking a Task
If task cannot proceed due to a dependency or external blocker:
Unblock when ready:
Task Lifecycle
| Transition | Trigger | Tool Call | |
|---|---|---|---|
| ready → claimed | Worker claims task | kanban_claim() | |
| claimed → in_progress | First heartbeat sent | kanban_heartbeat() | |
| in_progress → completed | Task done | kanban_complete() | |
| in_progress → blocked | Dependency unmet | kanban_block() | |
| blocked → in_progress | Blocker cleared | kanban_unblock() | |
| in_progress → failed | max_retries hit | kanban_complete() with failure |
Workspace
Workers operate in their own workspace:
| Field | Description | |||
|---|---|---|---|---|
workspace_kind | 'local' \ | 'docker' \ | 'ssh' | |
workspace_path | Path to working directory |
Use workspace_path as the working directory for all file operations.
Reference
See [[references/protocol.md]] for the full task lifecycle protocol, API schema, and examples.
Environment Variables
| Variable | Description | |
|---|---|---|
KANBAN_TASK_ID | Task ID assigned by dispatcher (if dispatched) | |
KANBAN_WORKER_PID | PID of the worker process | |
KANBAN_BOARD | Board name (default: 'default') | |
DREWENT_HOME | {{AGENT_NAME}} home path (~/.{{AGENT_NAME_LOWER}}) |
Dispatcher Notes (for cron job agents)
The kanban dispatcher is not invoked via from tools.{{AGENT_NAME_LOWER}}_kanban_db import dispatch_once — that import path does not resolve in production. The actual dispatch mechanism uses standalone scripts:
Scripts: ~/.{{AGENT_NAME_LOWER}}/scripts/dispatch_once_{board}.py (one per board: default, content, integrations)
Execution (run directly, no import gymnastics):
Each script opens its own SQLite connection and runs 5 phases:
- Phase 0 (watchdog):
os.kill(pid, 0)to check worker liveness; dead workers reclaimed immediately - Phase 0.5 (affinity): Skips tasks with 3+ consecutive failures (cooldown)
- Phase 1 (TTL reclaim): Reclaims tasks past
claim_expires(but skips if worker PID is still alive) - Phase 2 (claim): Claims ready tasks with adaptive
MAX_CLAIM(scales with queue depth) - Phase 3 (spawn): Spawns
run_kanban_worker.pyviaPopenwith logfile redirect (no PIPE deadlock)
Worker logs: ~/.{{AGENT_NAME_LOWER}}/P4-cortex/scripts/kanban/logs/workers/{task_id}.log
Output protocol: [SILENT] if no work done; otherwise detailed breakdown with per-task details.
Related Skills
- [[kanban-orchestrator]] — Task decomposition and linking
- [[kanban-dashboard]] — Board visualization via n8n
- [[kanban-dispatcher-hardening]] — Dispatcher watchdog + logfile redirect internals
- [[kanban-dispatcher-stalled]] — Diagnose dispatcher stalls when cron stops running
Related
- [[@memory/growth/{{AGENT_NAME_LOWER}}-kanban-implementation-plan]]
- [[@action/skills/SKILL-INDEX]]
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/kanban-worker/SKILL.md
Popularity
- Repository stars
- 3
- Repository forks
- 8
- 7-day trend score
- 0
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/humanerd-drew/opencode-drewgent/kanban-worker)