| name | spectre-plan-route |
| description | Internal semantic classifier for spectre-plan. Use only from Plan with a bounded routing input; never for planning, artifacts, telemetry, or user gates. |
| user-invocable | false |
plan-route
Purpose
Return a deterministic plan-routing/v1 decision to spectre-plan. Classify only; Plan owns workflow authority.
Inputs
- Immutable canonical Scope + bounded repository observations; observations describe work, not size.
- Mode initial, or observed with the prior decision + plan Routing Observations.
- Resume-only legacy size is validated and normalized once.
Working Set
- Plan supplies one bounded scan: implementation topology, uncertainty, evidence, protected boundaries, graph risk, shipped precedent, and user-owned decisions.
- If missing evidence alone would cross a costlier route, use exactly one bounded probe (
@spectre:finder/@spectre:patterns) for one repository question.
Outputs + DONE
Return schema plan-routing/v1; shape ATOMIC|DIRECT|STRUCTURED; uncertainty LOW|MODERATE|HIGH; evidence SUFFICIENT|PROBE_REQUIRED|PROBED; protected_boundaries[] {type, threatened_invariant, failure_mode}; task_graph_risk LOW|HIGH; design flag; size; route; rationale. Observed also returns regret, reason codes, and KEEP|RERUN_SMALLER|RERUN_LARGER.
DONE when fields validate; any probe is consumed; STRUCTURED names workstreams; HIGH graph risk names an implementation-graph failure; and only the table derives size/route.
Method / guardrails
| Semantic result | Size · route |
|---|
| ATOMIC + LOW + no changed protected boundary | XS · XS_DIRECT |
| ATOMIC + LOW + changed protected boundary OR DIRECT + LOW | S · S_DIRECT |
| ATOMIC/DIRECT + MODERATE/HIGH | M · M_REVIEWED_DIRECT |
| STRUCTURED + LOW/MODERATE + ordinary graph risk | L · L_STRUCTURED |
| STRUCTURED + HIGH uncertainty or HIGH task-graph risk | XL · XL_REVIEWED_STRUCTURED |
- STRUCTURED requires multiple independently implementable workstreams; render variants of one surface (layout, density, breakpoint, theme), dependencies, supporting artifacts, workflow/acceptance steps, and pilots are not workstreams.
- Two shipped instances of a change-shape make the next DIRECT; its layers are not workstreams; STRUCTURED requires a named delta beyond repetition.
- HIGH graph risk requires a credible implementation ordering/coordination/rollback failure; workflow gates/state transitions do not qualify.
- A protected boundary needs a concrete invariant + credible failure mode; it floors size at S but never creates structure or HIGH graph risk.
- Honor confirmed Scope assumptions. Missing paths/evidence permit at most the one probe; when Scope mandates an abstraction, spend it on whether it ships; classify the real delta. Only unresolved, approach-changing uncertainty affects size.
- Design authority requires an unresolved product, compatibility, destructive, migration/rollback, or architecture choice outside implementation discretion; size and routine placement never create it.
- Observed reports regret and KEEP|RERUN_SMALLER|RERUN_LARGER; never repeats work or removes artifacts.
- Normalize legacy reads once: MICRO→XS, LIGHT→S, STANDARD-DIRECT→M, STANDARD→L, COMPREHENSIVE→XL; legacy labels never decide.
- Never use file volume, dependency count, surface counts, or sensitive-domain keywords as size authority, and never let them create workstreams. Never plan, write artifacts, emit telemetry, or present gates.
Handoff
Return the record plus probe evidence. Plan owns persistence, explanation, orchestration, gates, and telemetry.
Escalate-If
- Canonical Scope is absent/conflicting, an enum is invalid, or a protected boundary lacks a predicate.
- A material decision needs user authority; report
design_authority_required, never decide it.
