@mindrian_os/install 1.13.0-beta.13 → 1.13.0-beta.16

package/commands/rooms.md

@@ -4,6 +4,7 @@ description: List, switch, or archive project rooms
 argument-hint: [list|switch|archive|park]
 body_shape: B (Semantic Tree)
 serves_jtbd: ["audit-room"]
+teaching: "When you have multiple venture rooms and need to switch, list, or archive them, /mos:rooms manages the registry. One person, many ventures."
 ui_reference: skills/ui-system/SKILL.md
 allowed-tools:
   - Read

package/commands/root-cause.md

@@ -2,6 +2,7 @@
 name: root-cause
 description: Trace root cause via 5-Whys, Fishbone, Fault Tree
 serves_jtbd: ["find-problem"]
+teaching: "When the symptom keeps coming back, /mos:root-cause traces it via 5-Whys, Fishbone, and Fault Tree. Treats the cause, not the recurrence."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["Root Cause Analysis"]

package/commands/rs-experts.md

@@ -3,6 +3,7 @@ name: rs-experts
 description: Resolve the expert network for a topic via Aura Cypher MATCH
 body_shape: D (Comparison Matrix)
 serves_jtbd: ["find-bottleneck", "connect-domains"]
+teaching: "When you need to know who in the world is working on a reverse salient you found, /mos:rs-experts resolves the expert network via Brain Cypher MATCH. Routes you to the people who already know."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["Reverse Salient Analysis"]

package/commands/rs-explain.md

@@ -3,6 +3,7 @@ name: rs-explain
 description: Bidirectional NL-Graph entry point. NL question to graph queries to Larry-voiced explanation.
 body_shape: E (Action Report)
 serves_jtbd: ["find-bottleneck"]
+teaching: "When you have a question about a Reverse Salient discovery, /mos:rs-explain takes natural language in and returns a Larry-voiced explanation grounded in the graph. Bidirectional NL to graph and back."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["Reverse Salient Analysis"]

package/commands/rs-fetch.md

@@ -3,6 +3,7 @@ name: rs-fetch
 description: Run the full Reverse Salient discovery pipeline for a topic
 body_shape: E (Action Report)
 serves_jtbd: ["find-bottleneck", "surface-contradiction"]
+teaching: "When you need the full Reverse Salient pipeline run on a topic, /mos:rs-fetch executes the discovery end-to-end: corpus, math, cross-domain match, thesis. The complete sweep, not a sample."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["Reverse Salient Analysis"]

package/commands/rs-thesis.md

@@ -3,6 +3,7 @@ name: rs-thesis
 description: Read the thesis for a prior Reverse Salient discovery
 body_shape: E (Action Report)
 serves_jtbd: ["find-bottleneck"]
+teaching: "When you ran a Reverse Salient discovery earlier and want the thesis read back, /mos:rs-thesis surfaces the analytic conclusion in plain language. Best for revisiting old findings before a meeting."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["Reverse Salient Analysis"]

package/commands/scenario-plan.md

@@ -2,6 +2,7 @@
 name: scenario-plan
 description: Build a 2x2 scenario matrix of plausible futures
 serves_jtbd: ["compare-options", "plan-execution"]
+teaching: "When the future could go two different ways on two key uncertainties, /mos:scenario-plan builds the 2x2 matrix and names each quadrant. Forces you to plan for the world you do not expect."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["Scenario Planning"]

package/commands/scheduled-tasks.md

@@ -3,6 +3,7 @@ name: scheduled-tasks
 description: Define Cowork scheduled tasks for the room
 body_shape: E (Action Report)
 serves_jtbd: ["plan-execution"]
+teaching: "When you want Cowork to run something on a schedule against this room, /mos:scheduled-tasks defines the recurring job. Best for nightly grant sweeps or weekly meeting digests."
 ui_reference: skills/ui-system/SKILL.md
 surface: cowork
 allowed-tools:

package/commands/score-innovation.md

@@ -2,6 +2,7 @@
 name: score-innovation
 description: Score cross-domain innovation via HSI
 serves_jtbd: ["compare-options", "validate-idea"]
+teaching: "When you are choosing between cross-domain innovation candidates, /mos:score-innovation runs HSI scoring to rank them by semantic surprise. The math reveals which idea is actually novel."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["HSI Semantic Surprise Analysis Assistant"]

package/commands/scout.md

@@ -3,6 +3,7 @@ name: scout
 description: Run sentinel scans across the room
 body_shape: E (Action Report)
 serves_jtbd: ["explore", "understand-market"]
+teaching: "When you want background scans running across the room without driving them yourself, /mos:scout fires the sentinel checks. The proactive layer, not the reactive one."
 ui_reference: skills/ui-system/SKILL.md
 allowed-tools:
   - Read

package/commands/setup.md

@@ -3,6 +3,7 @@ name: setup
 description: Configure optional integrations (Brain, Velma)
 argument-hint: [brain|velma|graph]
 serves_jtbd: ["explore"]
+teaching: "When you want to wire optional integrations like Brain or Velma, /mos:setup walks you through configuration. MindrianOS works without them; they make it work harder."
 allowed-tools:
   - Read
   - Write

package/commands/snapshot.md

@@ -3,6 +3,7 @@ name: snapshot
 description: Package a Data Room snapshot for sharing
 argument-hint: '[<room-path>] [--open]'
 serves_jtbd: ["prepare-pitch", "audit-room"]
+teaching: "When you need a frozen Data Room artifact to share with someone outside the team, /mos:snapshot packages everything into a portable bundle. Read-only by design."
 disable-model-invocation: true
 usage: /mos:snapshot [ROOM_PATH] [--output PATH] [--open]
 category: export

package/commands/speakers.md

@@ -4,6 +4,7 @@ description: Show who spoke in your meetings and their roles
 body_shape: C (Room Card)
 body_shape_detail: Each speaker as a card with role, expertise, meeting count
 serves_jtbd: ["file-meeting"]
+teaching: "When you have a meeting filed and want to know who said what, /mos:speakers shows the participants with their roles, attendance, and contribution patterns. The people layer of meeting intelligence."
 ui_reference: skills/ui-system/SKILL.md
 allowed-tools:
   - Read

package/commands/splash.md

@@ -3,6 +3,7 @@ name: splash
 description: Display the MindrianOS Mondrian banner
 body_shape: raw
 serves_jtbd: ["explore"]
+teaching: "When you want the MindrianOS Mondrian banner, /mos:splash displays it. Mostly decorative; useful for screenshots and demo openings."
 allowed-tools:
   - Bash
 ---

package/commands/status.md

@@ -4,6 +4,7 @@ description: Show governing thought per section + health glyphs
 argument-hint: "[section] [--stale-only]"
 body_shape: E (Action Report)
 serves_jtbd: ["audit-room", "explore"]
+teaching: "When you need a fast read on the room's current state, /mos:status shows the governing thought per section plus health glyphs. The 10-second status check."
 ui_reference: skills/ui-system/SKILL.md
 allowed-tools:
   - Bash(node scripts/mos-status.cjs:*)

package/commands/structure-argument.md

@@ -2,6 +2,7 @@
 name: structure-argument
 description: Structure an argument with Minto + SCQA + MECE
 serves_jtbd: ["validate-idea", "explore"]
+teaching: "When an argument is muddled and you cannot say why, /mos:structure-argument restructures it with Minto pyramid, SCQA, and MECE. The right structure usually surfaces the missing premise."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["The Pyramid Principle", "MECE (Mutually Exclusive, Collectively Exhaustive)"]

package/commands/suggest-next.md

@@ -2,6 +2,7 @@
 name: suggest-next
 description: Suggest the next move using the room graph
 serves_jtbd: ["plan-execution", "explore"]
+teaching: "When you finish a step and want Larry to recommend the next move, /mos:suggest-next reads the room graph and proposes 3-5 options with reasons. The Navigation Engine made visible."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: meta
 frameworks: []

package/commands/systems-thinking.md

@@ -2,6 +2,7 @@
 name: systems-thinking
 description: Map feedback loops, stocks, and flows
 serves_jtbd: ["find-bottleneck"]
+teaching: "When the dynamics matter more than the parts, /mos:systems-thinking maps feedback loops, stocks, and flows. Surfaces where the leverage actually lives in the system."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["Systems Thinking"]

package/commands/think-hats.md

@@ -2,6 +2,7 @@
 name: think-hats
 description: Rotate through De Bono's Six Thinking Hats
 serves_jtbd: ["explore", "compare-options"]
+teaching: "When the team keeps wearing the same hat and missing perspectives, /mos:think-hats rotates them through de Bono's six. The discomfort is the point; that is where the new thought lives."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["Six Thinking Hats"]

package/commands/update.md

@@ -3,6 +3,7 @@ name: update
 description: Check for MindrianOS updates and install via Claude Code's native plugin loader
 argument-hint: [check|reapply|force]
 serves_jtbd: ["audit-room"]
+teaching: "When you suspect MindrianOS has a newer version waiting, /mos:update checks and installs via Claude Code's native plugin loader. Stale plugins quietly diverge from the docs."
 allowed-tools:
   - Bash
   - Read

package/commands/user-needs.md

@@ -2,6 +2,7 @@
 name: user-needs
 description: Map user needs with importance vs satisfaction
 serves_jtbd: ["find-problem"]
+teaching: "When you need to map what users actually want versus what they say they want, /mos:user-needs plots importance against satisfaction. The gap is where the opportunity lives."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["Jobs to Be Done (JTBD)"]

package/commands/validate.md

@@ -2,6 +2,7 @@
 name: validate
 description: Validate ideas via importance-satisfaction scoring
 serves_jtbd: ["validate-idea"]
+teaching: "When an idea needs a real test before more investment, /mos:validate runs importance-satisfaction scoring against the customer segment. Validation is a measurement, not a feeling."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["Jobs to Be Done (JTBD)"]

package/commands/value-proposition.md

@@ -2,6 +2,7 @@
 name: validate-proposition
 description: Score your value proposition against 3 VP gates
 serves_jtbd: ["validate-idea", "prepare-pitch"]
+teaching: "When you have a value proposition but no proof it holds, /mos:validate-proposition scores it against the three PWS VP gates with sequential math. A clean gate failure beats a vague pass."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["PWS Value Proposition"]

package/commands/vault.md

@@ -5,6 +5,7 @@ argument-hint: '[<room-name>] [--path <dir>]'
 disable-model-invocation: true
 body_shape_overview: E (Mini Report)
 serves_jtbd: ["prepare-pitch"]
+teaching: "When you want the Data Room available in Obsidian for offline reading, /mos:vault exports it as a nested vault with wikilinks intact. Graph view comes free."
 ui_reference: skills/ui-system/SKILL.md
 allowed-tools:
   - Read

package/commands/visualize.md

@@ -4,6 +4,7 @@ description: Open room diagrams in the browser
 argument-hint: [structure|graph|chart]
 body_shape: D (Document View)
 serves_jtbd: ["audit-room", "prepare-pitch"]
+teaching: "When you want to see the room as diagrams in a browser, /mos:visualize opens the visual layer. Best when a stakeholder needs the picture, not the prose."
 ui_reference: skills/ui-system/SKILL.md
 allowed-tools:
   - Bash

package/commands/whitespace.md

@@ -3,6 +3,7 @@ name: whitespace
 description: Detect whitespace gaps in the room's coverage
 body_shape: varies
 serves_jtbd: ["connect-domains", "find-problem"]
+teaching: "When you suspect a gap exists in the room's coverage of a domain, /mos:whitespace runs HSI scoring across the artifact corpus to find under-explored zones. Best after the room has 20+ entries."
 # --- Phase 122 workflow-layer frontmatter ---
 kind: methodology
 frameworks: ["HSI Semantic Surprise Analysis Assistant"]

package/commands/wiki.md

@@ -3,6 +3,7 @@ name: wiki
 description: Open the Data Room wiki of room sections
 body_shape: D (Document View)
 serves_jtbd: ["audit-room", "prepare-pitch"]
+teaching: "When you want to read the Data Room as linked wiki pages, /mos:wiki opens the wiki view. Section by section, with cross-references rendered as hyperlinks."
 ui_reference: skills/ui-system/SKILL.md
 allowed-tools:
   - Bash

package/lib/brain/framework-chain-slice.cjs

@@ -0,0 +1,193 @@
+'use strict';
+// Phase 125-02 -- Brain Cypher slice query (parameterized 1-3 hop FEEDS_INTO; LIMIT 50).
+// CONTEXT.md Scope IN section B item 4 -- the LOCKED Cypher. Async wrapper over
+// brain-client.query that produces the framework_chain_hint shape Plan 03 stitches
+// into the packet's local_graph_summary.
+//
+// Canon Part 8: only $active_frameworks (array of generic framework name strings,
+// each passed through sanitizeCypherInput) + $max_hops (1|2|3 int) cross the wire.
+// No user content. No artifact text. No room identifiers in the query.
+//
+// Graceful degradation: any failure path (Brain unreachable, query throws, invalid
+// max_hops, empty active_frameworks, all-rejected-by-sanitizer, non-array result)
+// returns a degraded result envelope rather than throwing. Tier 0 stays functional.
+
+const crypto = require('node:crypto');
+const defaultBrainClient = require('../core/brain-client.cjs');
+
+// LOCKED Cypher template per CONTEXT.md Scope IN section B item 4.
+// READ-ONLY (MATCH/RETURN/LIMIT only -- no MERGE/CREATE/DELETE). The two $-bound
+// params ($active_frameworks, $max_hops) are the ONLY values that cross the wire;
+// every framework name in $active_frameworks is sanitized via the brain-client
+// sanitizeCypherInput whitelist before binding (Canon Part 8 invariant).
+const FRAMEWORK_CHAIN_SLICE_CYPHER =
+  'MATCH (f:Framework)-[r:FEEDS_INTO*1..$max_hops]->(g:Framework) ' +
+  'WHERE f.name IN $active_frameworks ' +
+  'RETURN f.name AS from, g.name AS to, ' +
+  '       r.confidence AS confidence, ' +
+  '       r.transform_description AS transform_description, ' +
+  '       length(r) AS hop_distance ' +
+  'LIMIT 50';
+
+function _nowIso() {
+  return new Date().toISOString();
+}
+
+function _sha256Hex(s) {
+  return crypto.createHash('sha256').update(String(s), 'utf8').digest('hex');
+}
+
+function _degraded(opts) {
+  const max_hops = (opts && (opts.max_hops === 1 || opts.max_hops === 2 || opts.max_hops === 3))
+    ? opts.max_hops
+    : 3;
+  return {
+    edges: [],
+    slice_scope: max_hops,
+    slice_rationale: (opts && opts.rationale) ? String(opts.rationale) : 'degraded',
+    brain_snapshot_id: null,
+    fetched_at: _nowIso(),
+  };
+}
+
+// Normalize result rows from brain-client.query. The shipped brain-client.query
+// returns { records: Array<row> } for the brain_query path; tests + the
+// CONTEXT.md spec describe the inner array of rows. Accept BOTH shapes so we
+// stay compatible whether the caller stubs query() with an Array directly OR
+// the live brain-client wraps in { records }. Any other shape degrades cleanly.
+function _unwrapRows(result) {
+  if (Array.isArray(result)) return result;
+  if (result && Array.isArray(result.records)) return result.records;
+  return null;
+}
+
+/**
+ * Fetch a parameterized 1-3 hop FEEDS_INTO slice from the live Brain.
+ *
+ * @param {object} opts
+ * @param {string[]} opts.active_frameworks - Array of framework name strings.
+ *   Empty array short-circuits to a degraded envelope.
+ * @param {1|2|3} opts.max_hops - Hop depth (must be exactly 1, 2, or 3).
+ *   Any other value short-circuits to a degraded envelope.
+ * @param {object} [opts.brainClient] - Test seam; defaults to live brain-client.
+ * @returns {Promise<{
+ *   edges: Array<{from:string|null, to:string|null, confidence:number|null,
+ *                 transform_description:string|null, hop_distance:number|null}>,
+ *   slice_scope: 1|2|3,
+ *   slice_rationale: string,
+ *   brain_snapshot_id: string|null,
+ *   fetched_at: string
+ * }>}
+ */
+async function fetchFrameworkChainSlice(opts) {
+  const o = opts || {};
+  const active_frameworks = Array.isArray(o.active_frameworks) ? o.active_frameworks : [];
+  const max_hops = (o.max_hops === 1 || o.max_hops === 2 || o.max_hops === 3) ? o.max_hops : null;
+  const brainClient = o.brainClient || defaultBrainClient;
+
+  // Gate 1: max_hops must be 1, 2, or 3. Anything else (0, 4, undefined,
+  // string, null) returns a degraded envelope with slice_scope defaulted to 3.
+  if (max_hops === null) {
+    return _degraded({ max_hops: 3, rationale: 'invalid_max_hops; expected 1|2|3' });
+  }
+
+  // Gate 2: empty active_frameworks short-circuits before any Brain contact.
+  // The Cypher would match nothing anyway; the degraded envelope is honest.
+  if (active_frameworks.length === 0) {
+    return _degraded({ max_hops: max_hops, rationale: 'empty active_frameworks; no slice fetched' });
+  }
+
+  // Gate 3: Brain reachability check before issuing the Cypher query. The
+  // live brain-client.isAvailable() is sync + cheap (an API-key resolve).
+  if (typeof brainClient.isAvailable === 'function' && !brainClient.isAvailable()) {
+    return _degraded({ max_hops: max_hops, rationale: 'brain_unreachable' });
+  }
+
+  // Canon Part 8 sanitization. brain-client._test.sanitizeCypherInput is the
+  // shipped whitelist (Phase 110-05 D-04 invariant). Each framework name is
+  // sanitized BEFORE binding; values that scrub down to empty are dropped.
+  // Fallback no-op only when the seam is genuinely absent (the live
+  // brain-client always exposes it; tests may stub a minimal client).
+  const sanitizer = (brainClient._test && typeof brainClient._test.sanitizeCypherInput === 'function')
+    ? brainClient._test.sanitizeCypherInput
+    : function (s) { return s; };
+  const sanitized = active_frameworks
+    .filter(function (n) { return typeof n === 'string' && n.length > 0; })
+    .map(function (n) { return sanitizer(n); })
+    .filter(function (n) { return typeof n === 'string' && n.length > 0; });
+
+  if (sanitized.length === 0) {
+    return _degraded({
+      max_hops: max_hops,
+      rationale: 'all_active_frameworks_rejected_by_sanitizer',
+    });
+  }
+
+  // Issue the parameterized Cypher. The ONLY $-bound values are the sanitized
+  // framework name array + the max_hops integer. The Cypher string itself is
+  // a frozen constant -- no template interpolation, no string concatenation
+  // with caller-derived data (Canon Part 8 enforced by inspection).
+  let result;
+  try {
+    result = await brainClient.query(FRAMEWORK_CHAIN_SLICE_CYPHER, {
+      active_frameworks: sanitized,
+      max_hops: max_hops,
+    });
+  } catch (e) {
+    const msg = (e && e.message) ? String(e.message) : 'unknown';
+    return _degraded({
+      max_hops: max_hops,
+      rationale: 'brain_query_failed: ' + msg.slice(0, 60),
+    });
+  }
+
+  const rows = _unwrapRows(result);
+  if (rows === null) {
+    return _degraded({ max_hops: max_hops, rationale: 'brain_returned_non_array' });
+  }
+
+  // Map raw rows to the framework_chain_hint edge shape. Null-tolerant per
+  // G-05: confidence and transform_description may be null on the live graph
+  // and we preserve null explicitly (no defaults). hop_distance comes from
+  // length(r) which is always a number when the row exists.
+  // Defensive: hard-cap at 50 even though the Cypher already enforces LIMIT 50.
+  const edges = rows.slice(0, 50).map(function (r) {
+    return {
+      from: (r && typeof r.from === 'string') ? r.from : null,
+      to: (r && typeof r.to === 'string') ? r.to : null,
+      confidence: (r && typeof r.confidence === 'number') ? r.confidence : null,
+      transform_description: (r && typeof r.transform_description === 'string')
+        ? r.transform_description
+        : null,
+      hop_distance: (r && typeof r.hop_distance === 'number') ? r.hop_distance : null,
+    };
+  });
+
+  // brain_snapshot_id: per CONTEXT.md Open Question #4, "Brain commit_id if
+  // exposed, else SHA of Cypher result." The shipped brain_query does NOT
+  // expose a commit_id; we derive a deterministic content hash of the raw
+  // rows so callers can detect slice changes without holding the rows
+  // themselves (Plan 03 + downstream packet validator never sees them).
+  const snapshotId = _sha256Hex(JSON.stringify(rows));
+
+  return {
+    edges: edges,
+    slice_scope: max_hops,
+    slice_rationale: 'brain_reachable; ' + edges.length + ' edges fetched; '
+      + sanitized.length + ' active_frameworks; max_hops=' + max_hops,
+    brain_snapshot_id: snapshotId,
+    fetched_at: _nowIso(),
+  };
+}
+
+module.exports = {
+  fetchFrameworkChainSlice: fetchFrameworkChainSlice,
+  FRAMEWORK_CHAIN_SLICE_CYPHER: FRAMEWORK_CHAIN_SLICE_CYPHER,
+  // Test seam -- helpers exposed for invariant tests; not part of the
+  // public API. Plan 03 (packet builder) consumes only the two above.
+  _test: {
+    _degraded: _degraded,
+    _sha256Hex: _sha256Hex,
+    _unwrapRows: _unwrapRows,
+  },
+};

package/lib/core/cache-prune.cjs

@@ -31,10 +31,33 @@
  *
  * If installed_plugins.json is unreadable (ENOENT, parse error, or no
  * mos@mindrian-marketplace entry), the function returns
- *   { kept: [], removed: [], skipped: true, reason: '...' }
+ *   { kept: [], removed: [], removedBackups: [], skipped: true, reason: '...' }
  * with NO mutation. Never guesses; never deletes anything in the absence
  * of an authoritative active-version answer.
  *
+ * Phase 126 Plan-06 extension -- STALE BACKUP DIR PRUNING.
+ * In addition to the marketplace-cache version dirs, this function ALSO
+ * prunes Phase 95.2's atomic-swap backup directories at
+ *   ~/.claude/plugins/mindrian-os.stale-<tag>-<timestamp>/
+ * that are older than MOS_CACHE_PRUNE_AGE_DAYS (default 30 days).
+ *
+ * Background: Phase 95.2 wraps `claude plugin update` in an atomic-swap
+ * recovery flow. Each recovery leaves a backup of the prior install at
+ * the sibling path above. By Phase 95.2 contract, backups are retained
+ * "indefinitely" (after 24h the user can delete manually). Long-running
+ * tester installs (Lawrence, Gary, etc) never see manual cleanup, so
+ * backups accumulate. 30 days is the next operational threshold beyond
+ * the 24h user-driven cleanup window.
+ *
+ * Pattern match is LITERAL prefix `mindrian-os.stale-` (regex
+ * /^mindrian-os\.stale-/). Does NOT match `mindrian-os/` (the live
+ * install) or unrelated siblings -- the period after `mindrian-os` is
+ * what gates the match.
+ *
+ * Window override: set process.env.MOS_CACHE_PRUNE_AGE_DAYS to any
+ * positive integer. Invalid values (non-numeric, negative) fall back
+ * to the 30-day default. v2 may move this to .mos/config.json.
+ *
  * Signature:
  *   pruneMarketplaceCache({
  *     home = os.homedir(),
@@ -42,14 +65,22 @@
  *     retainCount = 2,
  *     dryRun = false,
  *   } = {}) => {
- *     kept: string[],     // version basenames kept on disk (incl. active)
- *     removed: string[],  // version basenames removed (or would be, in dryRun)
- *     skipped: boolean,   // true if we declined to act (see `reason`)
- *     reason: string|null // human-readable explanation, or null on success
+ *     kept: string[],            // version basenames kept on disk (incl. active)
+ *     removed: string[],         // version basenames removed (or would be, in dryRun)
+ *     removedBackups: string[],  // absolute paths of stale backup dirs removed (Phase 126)
+ *     skipped: boolean,          // true if we declined the marketplace-cache prune
+ *     reason: string|null,       // human-readable explanation, or null on success
+ *     ageDays: number,           // the active stale-backup age window in days (Phase 126)
  *   }
  *
+ * Backward compat: the original Phase 123 consumer at scripts/doctor.cjs
+ * line ~2100 reads r.removed.length + r.kept only; both unchanged here.
+ * `removedBackups` and `ageDays` are ADDITIVE fields.
+ *
  * Canon Part 8 sanity check (cp.6): the forbidden-network-token grep
- * exits 1 (no match) on this source file. Test cp.6 enforces it.
+ * exits 1 (no match) on this source file. Test cp.6 enforces it. The
+ * Phase 126 extension reads only local fs (stat + readdir + rmSync on
+ * scratch / on-disk backup dirs); zero network surface.
  */
 
 const fs = require('node:fs');
@@ -137,10 +168,20 @@ function pruneMarketplaceCache(opts) {
   const retainCount = (typeof o.retainCount === 'number' && o.retainCount >= 0) ? o.retainCount : 2;
   const dryRun = !!o.dryRun;
 
+  // Phase 126 Plan-06: stale-backup age window. Read once at the top so the
+  // returned `ageDays` field reflects the value that gated the prune.
+  // Env override is integer-only; invalid values fall back to the default.
+  const ageDaysRaw = process.env.MOS_CACHE_PRUNE_AGE_DAYS;
+  const ageDays = (typeof ageDaysRaw === 'string' && /^\d+$/.test(ageDaysRaw))
+    ? parseInt(ageDaysRaw, 10)
+    : 30;
+
   // Step 1: read active version. Skip entirely if unavailable -- never guess.
+  // (When skipped, we also decline the Phase-126 stale-backup pass, mirroring
+  // "no mutation in the absence of authoritative state".)
   const active = readActiveVersion(home);
   if (active.skipped) {
-    return { kept: [], removed: [], skipped: true, reason: active.reason };
+    return { kept: [], removed: [], removedBackups: [], skipped: true, reason: active.reason, ageDays };
   }
   const activeVersion = active.activeVersion;
 
@@ -149,7 +190,9 @@ function pruneMarketplaceCache(opts) {
   if (names.length === 0) {
     // No cache dir to prune. The active version is still "kept" conceptually
     // (it lives in the cache when it exists; absent cache = nothing to do).
-    return { kept: [activeVersion], removed: [], skipped: false, reason: 'no cache dir to prune' };
+    // Phase 126: still run the stale-backup pass below (returns at end).
+    const removedBackups = pruneStaleBackups(home, ageDays, dryRun);
+    return { kept: [activeVersion], removed: [], removedBackups, skipped: false, reason: 'no cache dir to prune', ageDays };
   }
 
   // Step 3: build the keep-set.
@@ -191,18 +234,81 @@ function pruneMarketplaceCache(opts) {
     }
   }
 
+  // Phase 126 Plan-06: stale-backup pass. ADDITIVE to the cache prune above;
+  // walks PLUGIN_HOME (sibling of cache/) for mindrian-os.stale-* dirs older
+  // than `ageDays`. Failures are swallowed per-entry (best-effort, mirrors the
+  // existing per-version rmSync error path).
+  const removedBackups = pruneStaleBackups(home, ageDays, dryRun);
+
   return {
     kept: Array.from(keep),
     removed,
+    removedBackups,
     skipped: false,
     reason: null,
+    ageDays,
   };
 }
 
+// ---------------------------------------------------------------------------
+// Phase 126 Plan-06 helper -- prune stale backup directories.
+//
+// Walks <home>/.claude/plugins/ for entries whose basename matches the
+// LITERAL prefix /^mindrian-os\.stale-/ (the period after `mindrian-os`
+// is what gates the match -- does NOT match `mindrian-os/` live install
+// or unrelated siblings such as `mindrian-marketplace/`).
+//
+// Returns an array of absolute paths that were removed (or that would
+// have been removed in dryRun mode). On error (unreadable dir, stat
+// failure, rm failure), the entry is silently skipped -- best-effort
+// semantics mirror the in-loop pattern of the cache-version prune above.
+// ---------------------------------------------------------------------------
+function pruneStaleBackups(home, ageDays, dryRun) {
+  const removedBackups = [];
+  const pluginsDir = path.join(home, '.claude', 'plugins');
+  const cutoffMs = Date.now() - (ageDays * 86400000);
+
+  let entries = [];
+  try {
+    entries = fs.readdirSync(pluginsDir, { withFileTypes: true });
+  } catch (_) {
+    // pluginsDir absent: nothing to prune. Return empty list.
+    return removedBackups;
+  }
+
+  for (const e of entries) {
+    if (!e.isDirectory()) continue;
+    if (!/^mindrian-os\.stale-/.test(e.name)) continue;
+    const fullPath = path.join(pluginsDir, e.name);
+    let mtimeMs;
+    try {
+      mtimeMs = fs.statSync(fullPath).mtimeMs;
+    } catch (_) {
+      continue; // stat failure: leave for inspection
+    }
+    if (mtimeMs >= cutoffMs) continue; // recent enough; retain
+
+    if (dryRun) {
+      removedBackups.push(fullPath);
+      continue;
+    }
+    try {
+      fs.rmSync(fullPath, { recursive: true, force: true });
+      removedBackups.push(fullPath);
+    } catch (_) {
+      // Best-effort: a failed removal does not abort the whole prune.
+      // (Mirrors the existing pattern in the cache-version prune above.)
+    }
+  }
+
+  return removedBackups;
+}
+
 module.exports = {
   pruneMarketplaceCache,
   // Internal helpers exposed for testability + future reuse.
   readActiveVersion,
   listCacheVersions,
   sortByMtimeDesc,
+  pruneStaleBackups,
 };

package/lib/core/feynman/ROOM.md

@@ -0,0 +1,25 @@
+# lib/core/feynman/
+
+Phase 124 (FEYNMAN.md temporal awareness) renderer + runner ship here.
+
+Contains (after Plans 124-01 and 124-02 land):
+- `timeline-renderer.cjs` -- pure function `renderTimeline(db, sectionSlug, opts) -> { markdown_body, summary_stats }`. Reads ONLY via `lib/core/navigation.cjs` (the Phase 109 closed chokepoint): `findRecentChanges`, `findStaleDecisions`, and the new `firstCapturedLastTouchedBySection` (added as the 15th re-export in Plan 124-01, mirroring the Phase 110-03 `logMemoryEvent` re-export idiom). ZERO filesystem reads. ZERO Brain calls. ZERO LLM calls.
+- `timeline-runner.cjs` -- `refreshAll(roomDir)` + `refreshSection(roomDir, sectionSlug)`. Walks the room's section folders, finds each `FEYNMAN.md` with the sentinel pair (creates the pair if absent on first run), reads the surrounding body, calls the renderer, writes the file back with the body byte-preserved and the sentinel-bounded section replaced, sets `timeline_last_rendered: <ISO>` frontmatter (second-resolution). Each refresh logs a `memory_event` of type `feynman_timeline_refreshed`. Idempotent (re-run -> byte-identical output).
+
+Sentinel pair (D-02 hard invariant; bytes outside the pair are byte-preserved across regeneration):
+- `<!-- TIMELINE_AUTO_START -->`
+- `<!-- TIMELINE_AUTO_END -->`
+
+Stale thresholds (D-06; overridable via `process.env.MINDRIAN_TIMELINE_THRESHOLDS_JSON` for tests):
+- recent < 7 days
+- quiet 7-30 days
+- stale 30-90 days
+- dormant > 90 days
+
+Owner: Phase 124 FEYNMAN.md Temporal Awareness.
+Canon: Part 9 (the Larry-explains face of memory_event). Renderer reads ONLY room.db via navigation.cjs (D-03). Writes ONLY FEYNMAN.md inside the sentinels (D-02 hard invariant).
+Boundary: NO Brain calls (Canon Part 8); NO filesystem reads outside the FEYNMAN.md being written and the room.db family (allow-list pattern enforced by `tests/test-feynman-timeline-canon-part-9-invariant.cjs`).
+
+Upstream: `lib/core/navigation.cjs` (Phase 109 chokepoint), `lib/core/navigation/memory-events.cjs` (EVENT_TYPES +2 in Plan 124-02), `lib/core/navigation/insights.cjs` (source_section provenance).
+
+See: `.planning/phases/124-feynman-temporal-awareness/124-CONTEXT.md`.