agentleFS
zotoiozotoio/CRUX-Compress/.cursor/skills/crux-skill-memory-meditation-quick/SKILL.md

crux-skill-memory-meditation-quick

Skill8 starsChanged 2 months agoMITView on GitHub

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.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Verify flags.enableMemories is "true" (abort if not)

    line 21
  2. Distinctive:

    All four pre-spawn user gates run identically in Quick mode — Depth Selection (calling agent), Cost & Scope Acknowledgment (calling agent), …

    line 54
  3. Very specific:

    Verify flags.enableMemories is "true".

    line 58
  4. Very specific:

    Create meditations/{yyyymmdd}-{topic-slug}/.

    line 62
  5. Distinctive:

    No peer review in Quick mode.

    line 105
  6. Distinctive:

    No per-branch "Peer review" lines

    line 124
  7. Very specific:

    No Research-only facet-registry.yml / citations-index.yml lines under "Top-level artifacts"

    line 125

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

CRUX Skill: Memory Meditation — Quick Protocol

Implements the full Quick-mode meditation contract: 6-step parallel fan-out, step 4b 4-mode additional_focus_areas[] reconciliation, init-suggestions-{ts}.yml write, K10c in-pass reflection (same rubric as Research — warn-only citation regime carried through), comprehensiveness honouring at Quick leaf depth, and the Quick-mode steps 9–13 (identical to Research with documented relaxations).

When to Use

Load this skill when:

  • The crux-cursor-meditation-guide agent is spawned in Quick mode (--quick flag, meditateMode: "quick")
  • You are executing the Quick-mode depth-0 manager workflow (steps 1–13 with Quick relaxations)
  • You are executing as a Quick-mode child agent at depth 1, 2, or 3 (6-step protocol)

Never load this skill in Research mode, Ensemble Aggregation mode, or Adversarial Review function.

Prerequisites

  1. Verify flags.enableMemories is "true" (abort if not)
  2. Confirm comprehensiveness: payload is present — abort with: "comprehensiveness: payload required; missing from spawn prompt — caller misconfigured" if missing
  3. Confirm theming: payload is present — abort with a clear error pointing at Theme Preflight if missing
  4. Confirm modelStrategy: payload is present — abort with: "modelStrategy: payload required; missing from spawn prompt — caller misconfigured" if missing

Input Parameters (depth-0 manager, Quick mode)

Same shape as the Research depth-0 manager parameters (see the crux-skill-memory-meditation-research skill) with meditateMode: "quick". The only parameter differences are:

  • No facet-registry.yml operations (Quick mode skips global registry)
  • No citations-index.yml initialization (Quick mode does not maintain this file)

Input Parameters (Quick child agent, depth 1–3)

meditateMode: "quick"
workingDir: "<path>"
branchNumber: {N}
meditateDepth: {D}
maxDepth: {D}                     # propagated unchanged
subfocus: "<subfocus>"
subfocusSlug: "<kebab-case slug>"
subfocusIndex: {S}
parentSubfocus: "<parent's subfocus or null at depth 1>"
siblingFacets: [ ... ]
theming: { ... }                  # propagated unchanged — REQUIRED
comprehensiveness: { ... }        # propagated unchanged — REQUIRED; abort if missing
modelStrategy: { ... }            # propagated unchanged — REQUIRED; abort if missing
confirmDeepFacets: "none" | "depth_2_only" | "all_levels"
ensembleModel: "<model-slug>"     # when present, propagated unchanged; per-spawn model: carrier for this branch's subtree

Depth-0 Manager Workflow (Quick Mode)

All four pre-spawn user gates run identically in Quick mode — Depth Selection (calling agent), Cost & Scope Acknowledgment (calling agent), Theme Preflight (calling agent), and Facet Confirmation (subagent step 4 Pattern-B escalation). The post-consolidation steps 9–13 also run identically with the documented Quick-mode relaxations.

Step 1 — Check Feature Guard

Verify flags.enableMemories is "true". If not, return a message saying the feature is disabled and stop.

Step 2 — Create Working Directory

Create meditations/{yyyymmdd}-{topic-slug}/. If the directory already exists, append a numeric suffix.

Step 3 — SKIP

Quick mode does not use facet-registry.yml, citations-index.yml, or .facet-registry.lock/. Sibling-aware uniqueness uses facets.md only.

Step 4 — Derive Top-Level Facets + Draft Suggestions Payload and Confirm (Pattern B)

Derive the three facets AND draft suggestions payload (proposed_sections 3–8 items, proposed_visualisations 5–10 items, additional_focus_areas 0–5 items) as in Research. Facet descriptions are NOT required to carry citation backing at this stage; per-branch ## Citations requirements still apply to every child output. Run the identical combined Pattern-B flow (write extended facets-pending-{ts}.yml with all 4 blocks, escalate via combined needs_user_input block, resume with all 5 sub-question answers). Quick mode skips the facet-registry.yml append; the rest is identical including the source_signals discipline on every draft suggestion item.

The needs_user_input block schema is identical to the Research-mode block (see the crux-skill-memory-meditation-research skill, Step 4, for the full schema). Do NOT call AskQuestion — this is the calling agent's responsibility.

Step 4b — Resume-handler: apply confirmed payload, write init-suggestions-{ts}.yml, reconcile additional focus areas

Identical to Research with these Quick-specific differences:

  • facet-registry.yml operations are skipped (Quick mode only)
  • facets.md append still happens (the confirmed facet descriptions, citations, parent-context summary, and explicit partitioning statement)
  • init-suggestions-{ts}.yml is written with the same schema (see the crux-skill-memory-meditation-research skill for the full schema)

4-mode additional-focus-area reconciliation: same as Research — canonical additional_focus_areas[] array with per-item treatment: field:

  • skip → record with treatment: "skip", no facet, no section
  • additional_facet → new branch spawned alongside the original 3; no dedicated report section beyond natural branch output (at compact/default); or per-branch section at detailed/exhaustive
  • report_section_only → record with treatment: "report_section_only" and custom_report_section_title; populate resulting_section_id; report skill includes a section by that exact title
  • additional_facet_AND_section → both effects; populate both new_branch_index and custom_report_section_title

Cost-ack re-presentation: same rule as Research — if ANY focus-area decision is additional_facet OR additional_facet_AND_section, the calling agent fires the read-only-richness variant before this subagent resumes step 5. Only proceeds after cost_reack_confirmed: true.

Per-branch model assignment (modelStrategy.mode == "per_branch" only): identical to Research — after the final confirmed-branch count is known and any cost-ack re-presentation has resolved, resolve modelStrategy.branch_assignments[] per the policy in the /crux-meditate command's Argument Handling section (shuffle the pool with a topic+ts-derived seed; sample without replacement when poolSize ≥ branchCount, otherwise round-robin). The resolved assignments are surfaced in facets.md per the crux-skill-memory-meditation-coordination skill.

CRITICAL: additional_focus_areas_accepted and additional_focus_areas_skipped are LEGACY field names. The canonical schema uses a SINGLE additional_focus_areas[] array with per-item treatment: field. Never emit the legacy field names.

Hold onto confirmDeepFacets, comprehensiveness:, and modelStrategy: — all propagated unchanged to every child spawn in step 5.

Step 5 — Spawn Explorers

Identical to Research except each child receives meditateMode: "quick". The maxDepth, theming, comprehensiveness (propagated unchanged — abort if missing), modelStrategy (propagated unchanged — abort if missing), confirmDeepFacets, and ensembleModel (when present) payloads are all threaded through unchanged. Per-spawn model: dispatch is identical to Research — see the Research skill's Step 5 for the four-case rule (none / random / per_branch / ensemble_max). For per_branch, the depth-0 manager looks up modelStrategy.branch_assignments[branchNumber - 1].slug and sets it as both the child's model: and its ensembleModel so descendants inherit.

Step 6 — Poll for Branch Outputs

Identical poll loop (same prefix-glob, same resolve rule, same stale-lock guard is a no-op since no lock is ever created). The same deep-confirmation pending-facets-*.yml polling hook runs in Quick mode whenever confirmDeepFacets ≠ none.

Step 7 — SKIP

No peer review in Quick mode.

Step 8 — Consolidate + K10c Reflection

Consolidate from the branch files only at all depths that ran (per maxDepth). No peer-review files to glob, no citations-index.yml to merge. Follow the Subject-Matter Focus rule. If any branch surfaced citation gaps, include a "Citation gaps" callout in consolidation.md.

K10c in-pass reflection runs identically in Quick mode — same rubric, same scoring, same 5-candidate selection, same finalisation-enhancements.yml write. The only difference from Research is the set of inputs (no peer-review files, no citations-index.yml).

The finalisation-enhancements.yml schema is identical to Research except mode: "quick". See the crux-skill-memory-meditation-research skill for the full schema and K10c rubric.

Do NOT return to the calling agent yet — step 8b (K10b resume-handler) fires after the calling agent resolves Q-Finalisation-Enhancements, then steps 9–13 follow.

Step 8b — K10b Resume-handler for accepted enhancements

Identical to Research: re-read updated finalisation-enhancements.yml, process accepted cheap items into respawn payload, write follow-up artefacts for queued expensive items, accumulate pending_spawn_now list for spawn-now items.

Step 9 — Update facets.md with the Branch & Leaf Index (NOT SKIPPED)

Append the Branch & Leaf Index per the same format used in Research mode, with these omissions:

  • No per-branch "Peer review" lines
  • No Research-only facet-registry.yml / citations-index.yml lines under "Top-level artifacts"
  • Everything else (per-branch sections, depth-2/3 leaves, top-level consolidation.md + report pair + review iterations + confirmed-facets entries, missing-slots enumeration, index metadata) is identical

See the crux-skill-memory-meditation-coordination skill for the canonical Branch & Leaf Index template.

Step 10 — Adversarial Review and Fix Cycle (NOT SKIPPED)

Same reviewer agent, same iteration cap (3), same severity classification, same Pattern-B needs_user_input contract (with the mandatory context decision-guidance), same ESCALATE semantics, with two relaxations:

  • (a) "missing inline citation marker" findings are downgraded MUST_FIX → SHOULD_FIX (consistent with Quick mode's warn-only citation rule; unresolvable markers that do exist in the body remain MUST_FIX)
  • (b) the "peer review thoroughness" review dimension is N/A

Per-spawn model: dispatch for the reviewer is identical to Research — see the Research skill's Step 10. For modelStrategy.mode == "per_branch", omit model: so the reviewer runs on the caller's model (unified evaluator); for random pass the resolved slug; for none omit; for ensemble_max pass ensembleModel.

Reports are still gated on PASS / PASS_WITH_ADVISORIES; ESCALATE still aborts steps 11 and 12 exactly as in Research mode.

See the crux-skill-memory-meditation-review skill for the full review contract.

Step 11 — Re-run Branch & Leaf Index (NOT SKIPPED)

Identical to Research: refresh facets.md's Branch & Leaf Index after the reviewer's rewrites. Only runs when verdict from step 10 was PASS or PASS_WITH_ADVISORIES.

Step 12 — Generate the Mandatory Report (HTML + PDF)

Identical to Research per the fully-documented contract in the crux-skill-memory-meditation-report skill. The same theming, anti-homogenisation, Universal Contrast, light/dark, responsive-nav, PDF high-contrast print theme, clickable TOC, headless-Chrome render, filename-pairing rules, content-minimum requirements (level-driven per comprehensiveness), graceful PDF-degradation contracts, Report Comprehensiveness — No Information Loss contract, and Option Comparison Research mode all apply unchanged in Quick mode. Only runs when verdict from step 10 was PASS or PASS_WITH_ADVISORIES; never runs on ESCALATE.

Step 12b — Write Process Retrospective (NOT SKIPPED)

Identical to Research. Always written, including on ESCALATE.

Step 13 — Return to Calling Agent

Identical to Research: working directory path, facets.md, consolidation.md text + path, retrospective-{ts}.md path, report HTML+PDF pair (when generated), every review-pre-report-*-iter-*.md written by the review cycle. On ESCALATE return everything except report paths, plus a structured summary of unresolved MUST_FIX findings.

Quick Mode Child Protocol — 6-Step (Recursive, depth 1 through maxDepth - 1)

Each child agent at depth < maxDepth follows this simpler protocol.

1. Pre-derive 3 child subfocuses upfront (no prior research):
   Each must be narrower than this agent's subfocus, distinct from its siblings,
   and non-overlapping with the entries in facets.md. Sibling-aware only —
   Quick mode does NOT consult facet-registry.yml (the registry does not exist
   in Quick mode).

2. If confirmDeepFacets requires confirmation at this depth (depth_2_only at
   depth 1, OR all_levels at depth 1 or depth 2):
     - Write pending-facets-branch-{N}-depth-{D}-sub-{S}-{ts}.yml with the
       proposed 3 children (same schema as Research-mode Phase C; see the
       canonical Deep confirmation flow in the /crux-meditate command).
     - Poll for the matching confirmed-facets-...-{ts}.yml.
     - Apply per-child decisions (confirmed / modified / regenerate;
       regenerations capped at 3 per child).
   Only after all 3 children are confirmed, proceed to step 3.

3. Spawn 3 children at depth+1 in parallel with meditateMode: "quick".
   Each child receives meditateMode, workingDir, branchNumber, maxDepth,
   parentSubfocus, subfocus, subfocusSlug, subfocusIndex, siblingFacets, theming,
   comprehensiveness (propagated unchanged — MUST be present; child aborts with
   "`comprehensiveness:` payload required; missing from spawn prompt — caller
   misconfigured" if missing), modelStrategy (propagated unchanged — MUST be
   present; child aborts with the canonical error if missing), confirmDeepFacets
   (propagated unchanged), and ensembleModel (if present — propagated unchanged
   from this agent's own ensembleModel; descendants of a `per_branch` branch all
   inherit the SAME ensembleModel that was assigned at the depth-0 → depth-1
   dispatch point). When ensembleModel is set, pass model: ensembleModel on the
   Task tool invocation.

4. While children run, do this agent's own memory-query and expansion in
   parallel (Quick mode trades depth-first rigor for elapsed-time speed).

5. Wait for all 3 child files via prefix-glob
   branch-{N}-depth-{D+1}-sub-{S}-*.md. If a child's citations are missing or
   unresolvable, log a warning and proceed — do NOT respawn.

6. Aggregate children + own expansion into a single output file:
   branch-{N}-depth-{D}-sub-{S}-{slug}-{yyyymmddHHMMSS}.md
   (no rewrite — straight aggregation under ## Child Insights with
   `[child: branch-N-depth-D-sub-S]` provenance markers)

Leaf depth (deepest = maxDepth, Quick mode): query memories, expand, write the leaf file branch-{N}-depth-{D}-sub-{S}-{slug}-{ts}.md. Citations are still required in the file body and ## Citations section, but there is no parent respawn enforcement (warn-only). Comprehensiveness honouring in Quick leaf depth: at comprehensiveness.level ∈ {detailed, exhaustive} (depth3_leaf_inclusion = verbatim_quotes), leaf agents MUST include verbatim quoted passages from cited sources in their ## Discoveries section (same as Research mode). The section_length_budget_tokens.per_facet cap applies. Citation validation remains warn_only at every level — the parent does NOT respawn on citation failure in Quick mode regardless of comprehensiveness level (K7 mode-driven rule).

Output File Format (Quick branch files)

Identical frontmatter to Research branch files except mode: "quick". See the crux-skill-memory-meditation-research skill for the full frontmatter schema.

Body sections (mandatory in Quick mode):

  • ## Subfocus Rationale — why this narrowing was chosen
  • ## Discoveries — key findings from memory queries and research
  • ## Connections — patterns, relationships, non-obvious links
  • ## Child Subfocuses — the 3 narrower subfocuses derived upfront in step 1
  • ## Child Insights — aggregated from child output files in step 6 (straight aggregation with [child: branch-N-depth-D-sub-S] provenance markers — NOT a rewrite)
  • ## Contradictions — contradictions surfaced between this depth's findings and the children's
  • ## Summary — concise distillation for parent consumption
  • ## Citations — every source backing every claim. Mandatory in Quick mode; parent validates best-effort and surfaces gaps as warnings rather than re-spawning.

Citations Protocol (Quick Mode — Warn-Only)

Inline citation markers are the same vocabulary as Research ([memory: ...], [file: ...], [web: ...], [chat: ...], [child: ...]). Every output file requires a ## Citations section.

Validation rule — Quick mode: parents log warnings for missing or unresolvable citations and proceed (no respawn). The eventual report's executive summary must include a "Citation gaps" callout listing every uncited finding when this happens.

Quick mode does NOT maintain citations-index.yml — citation tracking is per-file only.

Quick vs Research Differences (Quick Column)

AspectQuick (this skill)
Recursion orderPre-derived: parent derives all 3 child subfocuses upfront, no prior research required
Facet uniquenessLocal sibling-aware only (read facets.md to avoid sibling overlap)
CitationsMandatory; inline markers + ## Citations section required, but parent validates best-effort and surfaces gaps as warnings rather than re-spawning
Bottom-up incorporationParent appends ## Child Insights section aggregating children (no rewrite)
Peer reviewNone
Consolidation inputsbranch-* files only
Coordination filesfacets.md only

For the Research column, see the crux-skill-memory-meditation-research skill.

Cross-Skill References

  • crux-skill-memory-meditation-coordination — filename grammar, polling rules, Branch & Leaf Index template
  • crux-skill-memory-meditation-review — adversarial review contract (step 10); Quick relaxations applied
  • crux-skill-memory-meditation-report — report generation contract (step 12)
  • crux-skill-memory-meditation-research — Research column of differences table + shared output schema reference + K10c rubric (same rubric; Quick uses warn-only citation regime)

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.