Skills model-queue

name: task-queue-by-model-source display_name: "Task Queue by Model Source" description: > Multi-queue task orchestration system. Tasks are routed to queues by model source, with support for task dependencies, context passing, and failure handling. Each model source has its own FIFO queue, executing one task at a time. version: 1.0.0 tier: general

metadata: author: openclaw created: "2026-03-04" updated: "2026-03-04"

modes:

• name: INTAKE trigger_type: user_message description: Parse user message into tasks, route to appropriate queue

• name: DISPATCHER trigger_type: heartbeat_or_cron description: Check queues, dispatch pending tasks, handle completions

triggers: positive: - "add task" - "new task" - "queue this" - "do this for me" - "handle this" - "task status" - "show queue" - "what's pending" - "cancel T-" - "retry T-" - "skip T-" - "用.*模型" - "send to"

system_events: - "MODEL_QUEUE_DISPATCH: check queues and run pending tasks"

configuration: tools_md_keys: MODEL_QUEUES_DIR: default: "~/.openclaw/model-queues/" type: path description: Directory for queue files and archives

MODEL_QUEUE_MAX_RETRIES:
  default: 3
  type: integer
  description: Max retry attempts per task

MODEL_QUEUE_ARCHIVE_DAYS:
  default: 7
  type: integer
  description: Days before archiving completed tasks

MODEL_SOURCE_*:
  type: mapping
  description: Model source definitions (e.g., MODEL_SOURCE_OLLAMA_LOCAL=ollama/llama3,ollama/qwen2.5)

task_states:

• pending # Ready to be dispatched
• waiting # Waiting for dependency
• running # Currently executing
• done # Completed successfully
• failed # Failed after max retries
• blocked # Dependency failed, needs action
• skipped # Skipped by user or dependency failure

dependency_handling: on_depends_fail: - block # Mark as blocked, wait for user action - skip # Skip the task automatically - continue # Execute anyway with warning context

outputs:

• name: queue_files type: file (JSON) path: "${MODEL_QUEUES_DIR}/{source}.json" description: One queue file per model source

• name: notifications type: message (chat) description: Task completion/failure notifications

• name: archive_files type: file (JSON) path: "${MODEL_QUEUES_DIR}/archive/{source}/{YYYY-MM}.json" description: Archived completed tasks

heartbeat_integration: register_in_heartbeat_md: true heartbeat_check: "Check all queue files in ${MODEL_QUEUES_DIR}; if any pending/running tasks exist, run DISPATCHER mode; else HEARTBEAT_OK" cron_backup: schedule: "every 15 minutes" system_event: "MODEL_QUEUE_DISPATCH: check queues and run pending tasks"

permissions: filesystem: - "create ${MODEL_QUEUES_DIR}/ directory on first run" - "read/write queue files in ${MODEL_QUEUES_DIR}/" - "create archive directories" - "update HEARTBEAT.md"

cron: - "register one backup cron job on first run"

subagents: - "spawn subagent per running task" - "check subagent session status"

dependencies: tools: required: - read - write - exec - sessions_spawn - subagents optional: - cron - message

references:

• queue-schema.md

first_run_behavior: > On first INTAKE (no queue files exist):

1. Create ${MODEL_QUEUES_DIR}/ directory
2. Create queue files for each configured model source
3. Register heartbeat entry in HEARTBEAT.md
4. Register backup cron job
5. Notify user: "⚙️ Model Queue initialized. N queues ready."

examples:

• user: "用 qwen2.5 分析一下销售数据" response: "📋 Added T-001 to queue [ollama-local]\nModel: ollama/qwen2.5\nTask: 分析销售数据\nQueue position: 1"

• user: "然后生成报告" response: "📋 Added T-002 to queue [ollama-local]\nModel: ollama/qwen2.5\nTask: 生成报告\nDepends on: T-001\nQueue position: 2"

• user: "任务状态" response: | 📊 Queue Status

  [ollama-local] 1 pending, 1 running, 0 done T-001 🔄 running: 分析销售数据 (started 2min ago) T-002 ⏳ waiting: 生成报告 (depends on T-001)