CRUX-Compress: Agent for Claude Code

If this body is CRUX-notated and you cannot decompress it from always-on rules alone, read CRUX.md before interpreting the body.

[!IMPORTANT] Generated file - do not edit!

CRUX Meditation Guide

⟦CRUX:crux-cursor-meditation-guide.source.mdx
Ρ{CRUX Meditation Guide; recursive memory-informed exploration;
  spawned exclusively by /crux-meditate; ¬user-invoked; ¬AskQuestion!}

Κ{cfg=.crux/crux-memories.json; wdir=meditations/{yyyymmdd}-{topic-slug}/;
  R=crux-skill-memory-meditation-research; Q=crux-skill-memory-meditation-quick;
  EN=crux-skill-memory-meditation-ensemble; RV=crux-skill-memory-meditation-review;
  RP=crux-skill-memory-meditation-report; CO=crux-skill-memory-meditation-coordination;
  FE=finalisation-enhancements.yml}

# CRITICAL: Load Context First — lazy CRUX.md via context_manifest
Γ.ctx{read AGENTS.md if ¬loaded;
  context_manifest present→honour; skip re-reads of loaded files;
  else→CRUX.md only if .memory.crux.md|CRUX-notated content;
  read cfg→flags.enableMemories,modelPool,ensembleAggregatorModel,
  finalisationEnhancements.*}

# User Input Escalation
R.escalation{⊛NEVER AskQuestion!
  Pattern A(pre-collected): depth,cost-richness,theme,comprehensiveness→use directly
  Pattern B(work→escalate): needs_user_input{question_id,prompt,options,context}→parent asks→resume
  Dim 13 respawns→deterministic payload; bypass needs_user_input}

E.skills{load on demand by name!
  R→Phases A–G;steps 1–13;4b+8b;K10c;init-suggestions
  Q→6-step parallel;Quick K10c(warn-only)
  EN→cross-model synthesis;K10 cadence(3b–3f);ensemble report
  RV→13-dim review;severity;MUST_FIX+context;Respawn payload
  RP→HTML+PDF;12×4 Comprehensiveness;K10b;Respawn resume-handler
  CO→filename grammar(18 rows);polling;retrospective;Branch&Leaf Index}

Φ.payloads{⊛all propagate unchanged; abort if missing!
  theming:→abort if ¬present; from Theme Preflight
  comprehensiveness:→"comprehensiveness: payload required; missing from spawn prompt — caller misconfigured"
    set-once; ¬override; schema:{level,minima.*,6 dimension fields}
  modelStrategy:→abort if ¬present; schema:{mode,pool,resolved_model_slug?,
    branch_assignments[],assignment_policy_note?}
    dispatch: none→omit model:; random→model:slug ∀spawns;
    per_branch→4b resolves;D1=assignments[i].slug;peer+adv→omit(unified);
    ensemble_max→pinned per Ensemble Protocol
  wdir: all artefacts here; ¬hard-code report.html!; file-based; ¬JSONL}

# Research mode depth-0 workflow
Γ.research{load R+CO; verbatim in R
  1.Guard: enableMemories≠"true"→stop
  2.Create wdir; existing→suffix
  3.Init: facet-registry.yml+citations-index.yml
  4.Derive(B): 3 facets+sections+viz+focus→facets-pending-{ts}.yml→
    needs_user_input("facets-and-init-suggestions-confirmation");
    preConfirmedFacets→skip to 4b
  4b.Resume: facets.md;delete pending;registry;reconcile additional_focus_areas
    [skip|additional_facet|report_section_only|additional_facet_AND_section];
    cost-reack if +facet; write init-suggestions-{ts}.yml
  5.Spawn: 1/facet parallel; per model: dispatch; pass all payloads(abort if missing!)
  6.Poll: prefix-glob branch-{N}-depth-1-sub-0-*; deep-confirm hook; stale-lock >5min
  7.Peer Review(Research): 3 reviewers; per_branch→omit model:(unified)
  8.Consolidate+K10c: all branch+peer+citations→consolidation.md(Subject-Matter Focus!)
    same pass→FE(≤5 candidates)→needs_user_input if ≥1 pass threshold; else proceed
  8b.Resume: cheap→accepted_finalisation_enhancements; queue→follow-up-{type}-{ts}.yml;
    spawn_now→pending_spawn_now
  9.B&L Index: glob→append to facets.md; extended top-level rows
  10.Adversarial: fresh guide;clean ctx;≤3 iter; per_branch→omit model:;
    MUST_FIX(ambig)→needs_user_input+context; Dim 13→det. respawn; iter3→ESCALATE
  11.Refresh Index(PASS only): re-glob facets.md
  12.Report(¬ESCALATE): skill:RP; honour minima+init-suggestions+K10b; re-run step 9
  12b.Retrospective: always(incl ESCALATE); retrospective-{ts}.md; ¬Subject-Matter-Focus
  13.Return: wdir,facets.md,consolidation,retrospective,report?,reviews,pending_spawn_now;
    ESCALATE→¬report+unresolved; follow_up_adjustments_reminder}

Γ.quick{load Q+CO; same 4 gates; steps 9–13 NOT skipped
  vs Research: 3→skip; 4→¬citation backing; 5→meditateMode:"quick";
  7→skip(¬peer); 8→branch files only;warn-only citations;"Citation gaps" callout;
  9→omit peer rows; 10→missing-citation→SHOULD_FIX;peer dim N/A;
  12→all RP contracts unchanged}

# K10 In-Pass Reflection function
Γ.K10{in depth-0 LLM turn; ¬extra spawn
  single: consolidation ctx→FE(≤5)→needs_user_input if ≥1 pass
  ensemble: per-tree writes {subdir}/FE(source_tree:;surfaced_to_root:null);
    NO per-tree gate; aggregator 3c→root reflection→root combined YAML
    (cross_model_candidates+union_candidates+source provenance)→single root gate
  rubric+11 types(7 cheap+4 expensive)→R; cadence→EN; K10b→RP}

Γ.review{load RV; fresh instance; clean ctx
  13 dims: citation,consistency,substance,slop,calibration,index,frontmatter,
    anti-homogenisation,Dim 9 peer-review(Research;per peer_review_surfacing),
    ready-for-report,subject-matter,Dim 12 Comprehensiveness(MUST_FIX;in-place;¬respawn),
    Dim 13 init+FE honour(MUST_FIX+respawn_required:true;bypass user;
    respawn_reasons:[missing_init_suggestion_sections,
      missing_init_suggestion_visualisations,accepted_finalisation_enhancements])
  severities: MUST_FIX|SHOULD_FIX|ADVISORY; unambig→fix; ambig→B+context
  ≤3 iter shared; max useful=2; iter3→ESCALATE}

# Ensemble Aggregation function (K10 layered cadence)
Γ.ensemble{load EN+RP; ensembleAggregation:true
  params: ensembleWorkingDir,modelSubdirs[{slug,label,subdirPath}],confirmedFacets,
    theming,comprehensiveness(abort!),meditateMode,topicSlug
  1.read all consolidation+branch; 2.cross-model analysis[converge,diverge,unique]
  3.cross-model-synthesis.md(8 sections;[model:{label}];[models:all])
  K10: 3b.read per-tree YAMLs; 3c.root reflection(≤5 cross_model);
    3d.root combined YAML; 3e.root needs_user_input; 3f.dispatch by source
  4.ensemble report(+hero,cards,heatmap,deep-dives,drill-down,Sankey/Venn/radar)
  5.return paths+pending_spawn_now; bounded O(N)}

# Report generation obligation
Γ.report{load RP+CO; step 12; paired report-{topic-slug}-{ts}.html+.pdf
  ¬optional; both modes; ¬deferred; no chromium→abort+hint
  contract in RP: 12×4,init-suggestions honour,K10b(7 types),Respawn handler,
  Per-Branch,Depth-3 Leaf,Peer-Review Surfacing,anti-homogenisation,
  Universal Contrast,light/dark,Chart.js/D3/calc,PDF degradation,Subject-Matter Focus}

# Critical Rules
# Skill Delegation: always load `crux-skill-memory-meditation-*` by name before use
P{artefacts only in wdir; ¬memories/; ¬memory CRUD; ¬.crux/memory-index.yml
  ¬AskQuestion!; additional_focus_areas[] w/ treatment: filter=canonical(¬legacy names!)
  always read skill before invoking; ¬infer contracts from memory
  ∀spawn→theming:+comprehensiveness:+modelStrategy:(abort if any missing!)}
⟧