CRUX-Compress: Skill for Cursor

CRUX Skill: Memory Meditation — Ensemble Aggregation

Implements the Ensemble Aggregation function: cross-model synthesis after N independent meditation trees have completed, K10 layered cadence (steps 3b–3f), ensemble report extras, and K10 Ensemble Respawn Targeting. This skill is never loaded by per-model tree agents — only by the aggregator spawn.

When to Use

Load this skill only when:

• The crux-cursor-meditation-guide agent is spawned with ensembleAggregation: true
• You are the cross-model aggregator spawned by the calling agent's Ensemble Protocol (step 8 of the /crux-meditate command)

This skill is modelStrategy.mode == "ensemble_max" only. The other two pool-using strategies — random and per_branch — are single-tree variants that do NOT spawn an aggregator and do NOT produce a cross-model-synthesis.md or ensemble report pair. They run the standard single-tree workflow with model dispatch driven by the modelStrategy: payload (see the /crux-meditate command's Model Strategy payload section).

Never load this skill in:

• Per-model tree Research or Quick mode workflows (per-model trees use the crux-skill-memory-meditation-research or crux-skill-memory-meditation-quick skill)
• Single-tree random runs (no aggregation needed — one model, one report)
• Single-tree per_branch runs (no aggregation needed — one tree with per-branch model attribution recorded in facets.md and the standard single-tree report)

Prerequisites

1. All N per-model meditation trees have completed (calling agent confirms before spawning the aggregator)
2. comprehensiveness: payload is present in the spawn prompt — abort with: "comprehensiveness: payload required; missing from spawn prompt — caller misconfigured" if missing
3. theming: payload is present — abort with a clear error if missing

Input Parameters (Ensemble Aggregation invocation)

ensembleAggregation: true
ensembleWorkingDir: "<absolute path to the ensemble root directory>"
modelSubdirs: 
  - slug: "gpt-5.5-medium"
    label: "GPT 5.5"
    subdirPath: "<absolute path to model-gpt-5.5/>"
  # ... one entry per model
confirmedFacets: 
  - index: 1
    title: "{facet-1-title}"
    subfocus: "{facet-1-subfocus}"
    slug: "{facet-1-slug}"
  # ... 3 facets (shared, user-confirmed)
theming: { ... }                  # shared Theme Preflight payload — REQUIRED
comprehensiveness: { ... }        # shared comprehensiveness payload — REQUIRED; abort if missing
meditateMode: "research" | "quick"
topicSlug: "{topic-slug}"         # for report filenames