CRUX-Compress

Build a prioritised memory index from all memory files and reference trackers. Use when rebuilding the memory index, after memory CRUD operations, or when agents need a sorted discovery list.

Consolidate, promote, demote, and archive memories based on strength, usage patterns, and type transition rules. Use when running REM sleep, rebalancing memory strength, detecting conflicts, or cleaning up orphaned trackers.

Multi-purpose utility for CRUX compression workflows. Provides token estimation and checksum calculation. Use when estimating tokens for compression, comparing file sizes, or getting checksums for sourceChecksum tracking.

File-based coordination primitives for meditation: artefact filename grammar (18 rows including `init-suggestions-{ts}.yml`, `finalisation-enhancements.yml`, per-tree ensemble variants, and 4 follow-up artefact types), placeholders, prefix-glob polling rule, never-hard-code invariant, retrospective template, Branch & Leaf Index template (single-model + ensemble variants). Use whenever the `crux-cursor-meditation-guide` agent reads, writes, or links a working-directory artefact.

Ensemble Aggregation function for meditation: reads N model consolidations, writes `cross-model-synthesis.md`, performs K10 root cross-model reflection (step 3c), manages K10 layered cadence steps 3b–3f (per-tree `finalisation-enhancements.yml` reads + root combined YAML write with `cross_model_candidates` + `union_candidates`), returns single combined `needs_user_input`, dispatches resume-handler by `source` provenance, and hands off to the report skill for ensemble HTML+PDF generation. Use when the `crux-cursor-meditation-guide` agent is spawned with `ensembleAggregation: true`.

Quick-mode meditation protocol: 6-step parallel fan-out with optional deep-confirm hook, warn-only citation validation, upfront child derivation, no peer review, K10c reflection (same rubric, warn-only at every richness level per K7). Use when the `crux-cursor-meditation-guide` agent runs the Quick depth-0 manager or any Quick-mode child agent.

Mandatory paired HTML+PDF report generation for meditation: Comprehensiveness Level Mapping (12 dimensions × 4 levels), anti-homogenisation rules, Universal Contrast, light/dark mode + print TOC, Chart.js / D3 / calculator content minima with static fallbacks, Per-Branch Section Rule, Depth-3 Leaf Inclusion Rule, Peer-Review Surfacing Rule, Init-Suggestions Honour rules, K10b Per-Cheap-Type Rendering Contract (7 cheap types), and Report-Skill Respawn Protocol resume-handler. Use when the `crux-cursor-meditation-guide` agent runs report generation (step 12), when the ensemble aggregator generates the ensemble-level report, or when the report skill is respawned via Dim 13.

Research-mode meditation protocol: Phases A–G depth-first recursion, depth-0 manager steps 1–13 (incl. step 4b 4-mode `additional_focus_areas[]` reconciliation + `init-suggestions-{ts}.yml` write; step 8 K10c reflection writing `finalisation-enhancements.yml`; step 8b respawn-payload prep), facet registry lock, citations index, peer review file spec, comprehensiveness honouring at leaf depth. Use when the `crux-cursor-meditation-guide` agent runs the depth-0 manager or any Research-mode child agent.

Adversarial Review function for meditation: 13-dimension audit (incl. citation integrity, slop detection, anti-homogenisation drift, level-conditional Dim 9 peer-review thoroughness, Dim 12 Comprehensiveness fidelity, Dim 13 Init-suggestion AND finalisation-enhancement honour), severity classification, ≤3-iteration loop, MUST_FIX `needs_user_input` schema with mandatory `context` decision-guidance, and Dim 13 `respawn_required: true` Report-Skill Respawn Protocol payload schema (K9 + K10b). Use when the `crux-cursor-meditation-guide` agent is spawned in Adversarial Review function (step 10).

CRUX-compress memory files with adaptive sizing and source archival. Handles compress, decompress, migration of existing memories, and maxMemorySize enforcement (in lines). Use when compressing or decompressing memory files, or migrating uncompressed memories after enabling compression.

Analyse execution artifacts from completed units of work, compare with existing memories, detect conflicts, and propose ranked candidate facts for memory creation. Use after a spec (or configured unitOfWork) completes to extract learnings, patterns, red flags, ideas, and goals.

Create, read, update, delete, and validate memory files with proper frontmatter management. Use when creating new memories, modifying existing ones, removing memories, or validating memory file structure.

Track which memories are referenced in agent output, manage .refs.yml tracker files, and keep strength counters in sync with memory frontmatter. Use when recording a memory reference, cleaning up orphaned trackers during REM sleep, or checking whether a memory should be promoted to a permanent rule.