Tool: reasoning_intent_router

Purpose

Translate normal-language Thinking Mode requests into a concrete ReasonKit Think execution plan.

Input

  • intent: free-form text, e.g. "step by step first, then explore options, verify claims, export audit"

Output

  • user_mode
  • internal_method
  • mode_rationale
  • safety_notes
  • estimated_cost
  • mode_contract
  • inferred_modes
  • prompt_plan
  • tool_plan
  • resource_plan
  • suggested_tool_calls (with argument templates)
  • notes

Behavior

  • Detects Auto, Quick, Explore, Map, Sketch, Test, support-layer, and legacy CoT/ToT/GoT wording.
  • Returns deduplicated recommended order and call templates.
  • Always includes reasoning://thinking-modes in the resource plan so clients can inspect the full mode contract.
  • For local feature/TODO triage requests, includes reasoning://workflows/feature-triage and reminds the caller to read files with host tools before invoking ReasonKit.

Mode Vocabulary

  • Auto: recommend run_thinking_mode; reasoning_autopilot remains a compatibility Auto flow.
  • Quick: concise linear audit.
  • Explore: branch, score, prune.
  • Map: graph dependencies, merge, distill.
  • Sketch: outline-first Skeleton-of-Thought workflow.
  • Test: claim verification, policy gates, route decision. This is not runtime test execution.

Filesystem Boundary

The router can recommend workflows, but it cannot inspect @paths. If the intent mentions a feature directory, TODO list, repository path, or local file, the AI client must first list/read those files with its own tools and pass the extracted source packet into run_thinking_mode.