AgentSkillsCN

route-tasks-by-policy

根据仓库中的 routing.json 策略，将任务路由至特定模型，并将每一步决策记录至 routing_ledger.md。当用户请求调度、编排、路由，或需要多模型路由时，可使用此工具。

SKILL.md
--- frontmatter
name: route-tasks-by-policy
description: Route tasks to specific models using the repository routing.json policy and log each decision to routing_ledger.md. Use when the user asks to dispatch, orchestrate, route, or when multi-model routing is required.

Route Tasks By Policy

Do this

  1. •Read routing.json from the repository root.
  2. •Classify the task by matching keywords (case-insensitive) for each route.
  3. •If no match, select default_model and set task_type to unknown_default.
  4. •Log the decision to routing_ledger.md before executing the task.
  5. •Execute the task with the chosen model(s) and synthesize results if multiple models are used.

Helper script:

scripts/route_task.py --prompt "<task>" [--log]

Log format

Use a single-line entry:

[timestamp] | TASK: <task_type> | MODEL: <model> | REASON: <reason>

Rules

  • •Explain why the model was chosen by citing the route reason.
  • •Do not override routing.json unless the user explicitly instructs it.
  • •Log before execution.
  • •Do not edit routing.json or routing_ledger.md unless asked.