Tool: run_skeleton_elaboration

Purpose

Elaborate Skeleton-of-Thought skeleton nodes into bounded scaffold nodes or a prompt pack. This is phase 2 of the SoT workflow.

Current implementation mode: heuristic_scaffold. model_backed mode fails closed until a real provider invocation binding exists.

Input

  • deliberation_id
  • phase_id: optional if the session has a latest non-stale SoT phase.
  • skeleton_node_ids: optional subset of phase skeletons.
  • elaboration_depth: optional, clamped to 1..8, default 3.
  • max_parallel: optional requested parallelism telemetry value.
  • mode: prompt_pack | heuristic_scaffold | model_backed.
  • score_after: default true.
  • checkpoint_after: default true.
  • merge_after: default false.

Output

  • status: PROMPT_PACK_READY, HEURISTIC_NODES_CREATED, MODEL_BACKED_UNAVAILABLE, PARTIALLY_COMPLETE, or COMPLETE.
  • elaboration_node_ids_by_skeleton: nodes created under each skeleton.
  • prompt_pack: prompts the calling agent can use with add_thought_node.
  • scored_node_ids: ranked ids when scoring is enabled.
  • checkpoint_id: checkpoint created when enabled.
  • merge_node_id: merge node created when merge_after=true.
  • limitations: explicit provider/heuristic limits.

Behavior

  • Heuristic mode creates scaffold nodes only; it does not claim semantic reasoning.
  • Prompt-pack mode returns prompts without creating graph nodes.
  • Model-backed mode returns MODEL_BACKED_UNAVAILABLE and creates no fake content.
  • Duplicate elaboration of the same phase is rejected in the MVP.
  • Merge mode merges elaboration nodes, then adds strict depends_on edges from the merge node to elaboration sources.

Errors

  • Unknown phase or skeleton node.
  • No skeleton nodes available.
  • Duplicate elaboration attempt.
  • Node or edge budget exhausted.