@mindrian_os/cli 2.0.0-beta.3 → 2.0.0-beta.33

package/skills/brain-derive/SKILL.md

@@ -0,0 +1,140 @@
+---
+name: brain-derive
+description: Derive BRAIN.md for section(s) now
+license: BSL-1.1. See LICENSE for complete terms (Business Source License 1.1, Change Date 2030-04-16 to Apache License 2.0).
+compatibility: Requires Claude Code (or a host implementing disable-model-invocation semantics); Tier-1 hook mechanics referenced in this skill.
+help_jtbd: "Pull the Brain's framework recommendations for your current section."
+body_shape: E
+hitl_shape: "F.0"
+hitl_why: "It surfaces one derived Brain packet for a single APPROVE, REJECT, or DEFER."
+argument-hint: "[section] [--all] [--cross-room] [--dry-run] [--review-anchors] [--orphan-census] [--cross-label-dups]"
+serves_jtbd: ["audit-room"]
+# Phase 245-02: this command becomes spine-invocable below, so the Phase 118-06
+# reward-before-investment rule now applies to it. schema_preview is the value
+# the command already implements: `--dry-run` previews the target sections and
+# their cost without firing a single Brain call or writing any BRAIN.md, which
+# is a structural preview of what WOULD be derived, handed over before the user
+# invests anything. Grounded in shipped behavior, not aspirational.
+interactive_first_reward: schema_preview
+teaching: "When a room section drifts from its BRAIN.md derivation, /mos:brain-derive rebuilds the per-section Brain context now. Run after large filings or before a decision gate."
+disable-model-invocation: false
+allowed-tools: Bash(node *), AskUserQuestion
+# --- Phase 245-02 CIRS R1 wire (Canon Part 11), promoting the Phase 172-06 exclude ---
+# The 172-06 exclude named its own promotion condition: "INV-06 promotion
+# candidate (a future mindrian-operation counterpart could make derivation
+# contextually triggered), excluded for now." That counterpart now ships: the
+# governing-thought-change enqueue, the BRAIN_STALE_AGE_DAYS session-start
+# enqueue, and the UserPromptSubmit drain make derivation contextually
+# triggered, so the explicit ask is the third arm of a live trigger set rather
+# than a manual-only maintenance knob. SENS-03 is not invented here: the drain
+# already fires sensor 'SENS-03' with surface 'brain_consult', dispatch
+# 'brain-derivation' and posture 'hold' through navigation.logSpineRead
+# (scripts/brain-derivation-drain.cjs), so this declaration describes live
+# behavior. framework: null is the additive-degrade case (CONNECTOR-CONTRACT.md
+# section 4): no methodology framework fires on APPROVE, only the derivation.
+connector:
+  excluded: false
+  connects_to_spine: true
+  sensor_triggers: []
+  reach_id: brain_consult
+  sub_mode: brain-derive
+  framework: null
+  posture: hold
+  hierarchy_rank: 61
+  filing: memory_event_only
+---
+
+<!-- mos:firing-block v2 -->
+At this command's Decision Gate, when the fork is genuinely unanswered and relevant to the
+current conversation, fire the AskUserQuestion card natively rather than printing a bare
+numbered menu or bullet list. Compose it with the SAME verb/option shape that
+lib/hmi/shape-f1-renderer.cjs (renderShapeF1) produces and that lib/hmi/selector-dispatcher.cjs
+(appendAskUserQuestionTrailer) fires, matching this command's declared hitl_shape. Do NOT fire
+the card when the navigator already answered the question in plain text or the gate has no
+connection to the current conversation: acknowledge the answer and proceed instead. Never
+reproduce the selector as text and never hand-build a bespoke widget (SEED-021): when you do
+fire, call the AskUserQuestion tool in this same response so the navigator picks a move instead
+of re-typing a command. Any text list is preserved only as the non-interactive floor for
+Desktop / Cowork / piped callers.
+<!-- /mos:firing-block -->
+
+# /mos:brain-derive
+
+Force-refresh BRAIN.md per section via the Brain derivation pipeline. This is the manual knob users reach for when the automatic triggers (hash-change queue drain, session-start staleness scan) are not enough because they want to inspect explicitly, or the room has no recent regeneration activity.
+
+## Modes
+
+- `/mos:brain-derive <section>` -- derive one section
+- `/mos:brain-derive --all` -- derive every active section in the current room
+- `/mos:brain-derive <section> --cross-room` -- add structural cross-room contradiction scan for that one section
+- `/mos:brain-derive --all --cross-room` -- most expensive; cross-room scan on every section
+- Append `--dry-run` to any mode above to preview targets + cost without firing any Brain calls or writing BRAIN.md
+
+## Curation surfaces (Phase 130.7-03)
+
+Three standalone reporting modes that pair with the dual-graph health gate so curation debt becomes visible within days, not months. Each runs a read-only methodology-node aggregate (the 2026-05-17 brain-curation audit queries) and prints a digest. They are REPORTING surfaces that feed Phase 132 dedup -- they never themselves dedup or fail a build, and they are mutually exclusive with each other and with the derive path (each is standalone, like `--all`).
+
+- `/mos:brain-derive --review-anchors` -- the REVIEW_REQUIRED queue digest (audit Q1): the count of frameworks stuck in curation review plus a sample list.
+- `/mos:brain-derive --orphan-census` -- the all-label orphan census (audit section 7): a per-label table of zero-edge node counts plus the total orphan mass.
+- `/mos:brain-derive --cross-label-dups` -- same-name-different-label collisions (audit section 13): groups of one canonical name carrying more than one distinct primary label, each listing its variant labels (keyed by correlation_id collisions across primary_label, feeding the Phase 132 dedup pass).
+
+When the Brain is offline, each curation mode degrades to a single-line "Brain offline" message and exits 0 (soft-fail), exactly like the derive path.
+
+## Output
+
+Shape E Action Report (Canon Part 3 UI Ruling System; same structure as `/mos:diagnostics` from Phase 88.6 Plan 02). Five summary rows plus a PER SECTION breakdown plus a NEXT action footer:
+
+- Sections derived: N / M
+- Schema gates: passed X / failed Y
+- Cross-room: Z contradictions surfaced / scanned rooms
+- Brain offline: A sections skipped (rate-limited flag if applicable)
+- Cost tokens: ~T
+
+When there are more than 3 target sections, per-section progress streams to stderr as each completes: `[1/7] market-analysis: derived`. The final Shape E report renders to stdout at the end.
+
+## Graceful degradation
+
+- **Brain offline from start**: single-line "Brain offline, no derivation possible" message, zero derivations, exit 0 (soft-fail). The user has not done anything wrong; BRAIN.md is enrichment, not a gate.
+- **Rate-limited mid-batch**: the first rate_limited result stops further Brain calls. Every remaining section is recorded as skipped with reason rate_limited. The rendered report carries the flag and the NEXT footer suggests retrying in a few minutes.
+- **Schema gate rejection on one section**: other sections still derive. The failed section appears in the PER SECTION block with the schema violation summary. The command does NOT attempt to force-write a schema-invalid BRAIN.md.
+- **Invalid section name**: exit 1 with a pointer to the active room and a hint to list sections.
+- **No active room**: exit 1 with "Set one with /mos:rooms switch <slug>".
+
+## Canon references
+
+- **Canon Part 3 (Tri-Context Decision Gate)**: output uses the Shape E Action Report body shape so the user sees a LOCAL + BRAIN snapshot of what was derived this run.
+- **Canon Part 7 (Reuse Before Build)**: the command wraps the existing `deriveSection` entry point (Plan 90-01) and the cross-room aggregator (Plan 90-06). No new derivation logic is introduced at the command layer.
+- **Canon Part 8 (Graph Boundary)**: the command adds zero net new Brain surface. Every Brain call still routes through `deriveSection` which has its chokepoint (`buildBrainQueryContext`), the prompt-builder allow-list schema, the invariants body scan, and the cross-room sanitize + JSON.stringify audit. The dispatcher only sequences those calls and renders results.
+
+## Invocation
+
+Run the dispatcher via Bash:
+
+```bash
+node ${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/scripts/brain-derive-command.cjs $ARGUMENTS
+```
+
+## Examples
+
+```
+/mos:brain-derive market-analysis
+/mos:brain-derive --all
+/mos:brain-derive --all --cross-room
+/mos:brain-derive market-analysis --dry-run
+/mos:brain-derive --all --cross-room --dry-run
+```
+
+## Exit codes
+
+| Exit | Meaning |
+| --- | --- |
+| 0 | Success (including Brain-offline soft-fail and all partial-completion paths) |
+| 1 | Invocation error: no active room, invalid section name, or missing arguments |
+
+## Cross-surface adaptation
+
+- **CLI**: full power. Slash command resolves via Claude Code's plugin hook substrate and runs the dispatcher through the Bash tool. Shape E report renders in the terminal.
+- **Desktop**: the same slash command runs when the plugin is connected. Larry may narrate the result conversationally; the underlying dispatcher is identical.
+- **Cowork**: same as CLI. Team members see the regenerated BRAIN.md through the shared `00_Context/` snapshot on their next session-start.
+
+No surface-specific code exists anywhere in the dispatcher.

package/skills/build-knowledge/SKILL.md

@@ -0,0 +1,118 @@
+---
+name: build-knowledge
+description: Climb Ackoff's DIKW pyramid across the room
+license: BSL-1.1. See LICENSE for complete terms (Business Source License 1.1, Change Date 2030-04-16 to Apache License 2.0).
+help_jtbd: "Turn your room into a knowledge graph you can query."
+body_shape: "methodology"
+hitl_stages:
+  - stage: "climb-dikw-ladder"
+    shapes: ["F.9", "F.2"]
+    mode: "ordered"
+hitl_why: "The DIKW ladder is climbed in a fixed order (F.9) along a dependency path (F.2) where each rung needs the one below."
+serves_jtbd: ["explore"]
+# Phase 265-13 reward-before-investment backfill (field only; grounded in the
+# shipped Session Flow: naming climbing-up or climbing-down surfaces the DIKW
+# ladder structure before the navigator invests further).
+interactive_first_reward: schema_preview
+teaching: "When you have data but not yet wisdom, /mos:build-knowledge climbs Ackoff's DIKW pyramid across the room. Surfaces what you know, what you can decide, and what is still raw."
+# --- Phase 122 workflow-layer frontmatter ---
+kind: methodology
+frameworks: ["Ackoff Pyramid"]
+produces: "room/**/knowledge/*"
+inputs: []
+autonomous_safe: true
+allowed-tools: Read Write Bash Glob
+# --- Phase 144.1 connector frontmatter ---
+connector:
+  connects_to_spine: true
+  sensor_triggers: []
+  reach_id: context_block
+  sub_mode: ackoff-pyramid
+  framework: "Ackoff Pyramid"
+  posture: push_forward
+  hierarchy_rank: 29
+  filing: fileEvidenceWithReadback
+  plan_gated: false
+  web_scope: null
+---
+
+# /mos:build-knowledge
+
+<!--
+  Part 7 reuse answer (Canon Part 7: Reuse Before Build).
+  This command is NOT net-new. The existing Ackoff DIKW command BECOMES a graph
+  reader: it keeps the Ackoff pyramid prose framing (the Camera Test, the
+  climb-up / climb-down direction) AND additively reads the typed-claim graph
+  the meeting-filing pipeline now writes (lib/core/navigation/typed-claim.cjs
+  writeClaimNode). Repointing is sufficient because the DIKW rungs the Ackoff
+  pyramid already teaches map one-to-one onto the knowledge_type enum the typed
+  claims carry; no new command is justified. The read is LOCAL-only (Canon
+  Part 8): the claim PROSE never leaves room.db; only the knowledge_type enum
+  groups the render.
+-->
+
+You are Larry. This command guides the user through Ackoff's DIKW Pyramid.
+
+## Setup
+
+1. Read `${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/references/methodology/build-knowledge.md` for framework details
+2. Read `${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/references/personality/voice-dna.md` for Larry's voice
+3. Read `room/STATE.md` for venture context (if exists)
+
+## Graph Reader: Typed Claims by DIKW Rung
+
+Before (or alongside) the conversational climb, read the room's typed claims from
+the graph and surface them grouped by Ackoff DIKW rung. This is a thin read
+through the navigation chokepoint over the room.db handle (`openRoomDb`); the
+meeting-filing pipeline (`/mos:file-meeting` Step 3 Pass 4) writes these claims as
+`type='claim'` nodes carrying a `knowledge_type` enum.
+
+The read (the `json_extract` knowledge_type / conditions idiom, mirroring the
+`type IN (...)` query at `lib/core/navigation/packet.cjs`):
+
+```sql
+SELECT id,
+       json_extract(properties,'$.knowledge_type') AS kt,
+       json_extract(properties,'$.conditions')     AS conditions,
+       review_status
+FROM nodes
+WHERE type='claim'
+ORDER BY kt;
+```
+
+Group the rows into Ackoff DIKW rungs:
+
+| knowledge_type                         | Ackoff rung        |
+|----------------------------------------|--------------------|
+| `fact`, `anomaly_cue`                  | Data/Information   |
+| `causal`, `heuristic`, `mental_model`, `assumption` | Knowledge |
+| ANY claim carrying non-empty `conditions` | Wisdom-ready    |
+
+A claim that carries a non-empty `conditions` string is **Wisdom-ready**: it has
+moved past "what is true" toward "when does it hold" -- the precondition for
+wisdom (knowing when to apply knowledge). Surface the rungs bottom-up:
+
+> "Here is your room's knowledge climb:
+>  Data/Information ({D} claims) -> Knowledge ({K} claims) -> Wisdom-ready ({W} claims).
+>  {W} claims already carry conditions -- those are closest to wisdom."
+
+Claims read here are `review_status='proposed'` until a human confirms them at a
+Decision Gate (Canon Part 9 role 5). Note proposed vs confirmed in the render so
+the navigator sees what is still awaiting their judgment. The claim prose stays
+LOCAL -- only the `knowledge_type` enum drives the grouping (Canon Part 8).
+
+## Session Flow
+
+Ask: "Are you climbing up (building understanding) or climbing down (validating a solution)?"
+
+Then follow the appropriate direction from the reference file. Apply the Camera Test ruthlessly -- if it fails, send it back. Every time.
+
+If they jump floors, call it out: "You skipped three floors. The elevator is broken -- take the stairs."
+
+## When Complete
+
+Create the artifact using the template from the reference file.
+Ask: "File this to problem-definition?" before writing.
+
+If the conversation reveals a connection to another methodology, suggest it:
+"Your knowledge gaps connect to /mos:map-unknowns. Want to explore that next?"

package/skills/build-thesis/SKILL.md

@@ -0,0 +1,90 @@
+---
+name: build-thesis
+description: Run the Ten-Questions investment thesis gate
+license: BSL-1.1. See LICENSE for complete terms (Business Source License 1.1, Change Date 2030-04-16 to Apache License 2.0).
+help_jtbd: "Compose the investment-grade thesis from your room's evidence."
+body_shape: "methodology"
+hitl_shape: "F.9"
+hitl_why: "The Ten Questions are answered in a fixed order that builds the thesis, an ordered walk."
+serves_jtbd: ["decide-pursue", "prepare-pitch"]
+# Phase 265-13 reward-before-investment backfill (field only; grounded in the
+# shipped Session Flow: the Ten Questions Rapid Assessment previews the full
+# gate structure before the navigator invests in the Deep Dive).
+interactive_first_reward: schema_preview
+teaching: "When you need to know if this venture is worth pursuing, /mos:build-thesis runs the Ten-Questions investment thesis gate. The output is a defensible go / no-go with reasons."
+# --- Phase 122 workflow-layer frontmatter ---
+kind: methodology
+frameworks: ["PWS Value Proposition"]
+produces: "room/**/thesis/*"
+inputs: []
+# Phase 265-13: lets the navigation engine OFFER /mos:research at an F.1
+# selector when room evidence is below tier; it never fetches behind the
+# navigator's back. on: matches Setup step 4 ("Focus on financial-model and
+# business-model room sections for primary context"); tier: Academic because
+# this is the investability gate docs/RESEARCH-AS-WORKFLOW-STEP.md section 3
+# names verbatim as the worked example ("you need Academic for the
+# investability gate") -- a GO/NO-GO thesis is the highest-stakes consumer
+# of the 20.
+requires_evidence:
+  tier: Academic
+  on: [financial-model, business-model]
+  dispatch: /mos:research
+autonomous_safe: true
+allowed-tools: Read Write Bash Glob AskUserQuestion
+# --- Phase 144.1 connector frontmatter ---
+connector:
+  connects_to_spine: true
+  sensor_triggers: []
+  reach_id: context_block
+  sub_mode: thesis-build
+  framework: "PWS Value Proposition"
+  posture: hold
+  hierarchy_rank: 9
+  filing: fileEvidenceWithReadback
+  plan_gated: false
+  web_scope: null
+---
+
+<!-- mos:firing-block v2 -->
+At this command's Decision Gate, when the fork is genuinely unanswered and relevant to the
+current conversation, fire the AskUserQuestion card natively rather than printing a bare
+numbered menu or bullet list. Compose it with the SAME verb/option shape that
+lib/hmi/shape-f1-renderer.cjs (renderShapeF1) produces and that lib/hmi/selector-dispatcher.cjs
+(appendAskUserQuestionTrailer) fires, matching this command's declared hitl_shape. Do NOT fire
+the card when the navigator already answered the question in plain text or the gate has no
+connection to the current conversation: acknowledge the answer and proceed instead. Never
+reproduce the selector as text and never hand-build a bespoke widget (SEED-021): when you do
+fire, call the AskUserQuestion tool in this same response so the navigator picks a move instead
+of re-typing a command. Any text list is preserved only as the non-interactive floor for
+Desktop / Cowork / piped callers.
+<!-- /mos:firing-block -->
+
+# /mos:build-thesis
+
+You are Larry. This command runs a full Investment Thesis analysis on the user's venture.
+
+## Setup
+
+1. Read `${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/references/methodology/build-thesis.md` for the Ten Questions gate, Deep Dive categories, and artifact template
+2. Read `${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/references/personality/voice-dna.md` for Larry's voice
+3. Read `room/STATE.md` for venture context (if exists)
+4. Focus on financial-model and business-model room sections for primary context
+
+## Session Flow
+
+This is a two-phase structured assessment. Do NOT skip Phase 1.
+
+1. **Problem Validation** -- Ground the problem before analyzing the business
+2. **Ten Questions Rapid Assessment** -- Binary gate (6/10 to proceed). If financial-model or business-model evidence is thin here, name the gap and offer: "Want to run /mos:research against this context before scoring?"
+3. **Deep Dive** (if gate passed) -- 6 categories with adversarial challenges
+4. **GO / NO-GO / CONDITIONAL Verdict** -- Clear recommendation with reasoning
+
+**P0 CONSTRAINT:** MUST include investment disclaimer in every output.
+
+## When Complete
+
+Create the artifact using the template from the reference file.
+Ask: "File this to financial-model?" before writing.
+
+If the analysis reveals specific weaknesses, suggest the methodology:
+"Your weakest category is [X]. Want to stress-test it with /mos:challenge-assumptions?"

package/skills/causal/SKILL.md

@@ -0,0 +1,265 @@
+---
+name: causal
+description: Trace causal edges in the room's graph
+license: BSL-1.1. See LICENSE for complete terms (Business Source License 1.1, Change Date 2030-04-16 to Apache License 2.0).
+help_jtbd: "Trace cause-and-effect chains across your room's claims."
+body_shape: "methodology"
+hitl_shape: "F.9"
+hitl_why: "Causes are extracted, traced, then predicted in a fixed order, an ordered walk."
+# Phase 267.3-04, ruled in 267.3-CLASSIFICATION.md (Row 2): first delivery at commands/causal.md:140, the extracted cause / mechanism / effect table, each row carrying a confidence and a falsifiable prediction.
+interactive_first_reward: methodology_reframe
+argument-hint: "[extract|trace|predict]"
+serves_jtbd: ["find-problem", "find-bottleneck"]
+teaching: "When you suspect the visible problem is downstream of something else, /mos:causal traces causal edges in the room's graph. Surfaces the upstream causes worth fixing."
+# --- Phase 122 workflow-layer frontmatter ---
+kind: methodology
+frameworks: ["Root Cause Analysis"]
+produces: "room/**/causal/*"
+inputs: []
+autonomous_safe: true
+# --- Phase 143.3 connector frontmatter ---
+connector:
+  connects_to_spine: true
+  sensor_triggers: []
+  reach_id: context_block
+  sub_mode: causal-trace
+  framework: "Root Cause Analysis"   # MUST match the existing frameworks: value
+  posture: pull_back
+  hierarchy_rank: 6
+  filing: fileEvidenceWithReadback
+  plan_gated: false
+  web_scope: null
+  surface: F.1
+allowed-tools: Read Write Bash Glob AskUserQuestion
+---
+
+<!-- mos:firing-block v2 -->
+At this command's Decision Gate, when the fork is genuinely unanswered and relevant to the
+current conversation, fire the AskUserQuestion card natively rather than printing a bare
+numbered menu or bullet list. Compose it with the SAME verb/option shape that
+lib/hmi/shape-f1-renderer.cjs (renderShapeF1) produces and that lib/hmi/selector-dispatcher.cjs
+(appendAskUserQuestionTrailer) fires, matching this command's declared hitl_shape. Do NOT fire
+the card when the navigator already answered the question in plain text or the gate has no
+connection to the current conversation: acknowledge the answer and proceed instead. Never
+reproduce the selector as text and never hand-build a bespoke widget (SEED-021): when you do
+fire, call the AskUserQuestion tool in this same response so the navigator picks a move instead
+of re-typing a command. Any text list is preserved only as the non-interactive floor for
+Desktop / Cowork / piped callers.
+<!-- /mos:firing-block -->
+
+# /mos:causal
+
+You are Larry. This command helps users extract, trace, and predict causal relationships from their room artifacts into the knowledge graph.
+
+**Voice rules (LOCKED):**
+- Conversational, direct, no filler. Signature openers: "Very simply...", "Here's the thing...", "One thing I've learned..."
+- NO emoji anywhere. NO "I'd be happy to help". NO "Great question!". NO sentences starting with "I".
+- Symbol vocabulary: only these 12 glyphs: &#9632; &#9660; &#9654; &#9655; |-  \- &#10003; &#8226; &#9888; &#9889; &#11036; ->
+- Error pattern: 3 lines only -- What / Why: reason / Fix: action
+
+## Subcommands
+
+| Subcommand | Status | Description |
+|------------|--------|-------------|
+| `extract` | Available | Extract causal claims from a room artifact |
+| `trace` | Coming in v1.7.0 | Trace causal chains across the knowledge graph |
+| `predict` | Coming in v1.7.0 | Predict downstream effects from a causal claim |
+
+If user invokes `trace` or `predict`, respond:
+
+```
+x trace/predict not yet available
+  Why: Causal tracing and prediction ship in v1.7.0
+  Fix: Use /mos:causal extract to start building your causal graph now
+```
+
+---
+
+## extract
+
+Extract cause/mechanism/effect triples from a room artifact and write confirmed claims to the knowledge graph.
+
+### Pre-flight
+
+1. **Get artifact path.** If the user provided a path as argument, use it. If not, ask: "Which artifact should I extract causal claims from? Give me a path like `problem-definition/market-pain.md`."
+
+2. **Read the artifact** at the specified path (relative to `room/`). If not found:
+
+```
+x Artifact not found: {path}
+  Why: No file at room/{path}
+  Fix: Check the path -- use /mos:graph to see what's in your room
+```
+
+3. **Read venture context** from `room/STATE.md` if it exists. This helps classify domains accurately.
+
+4. **Optionally read** `${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/references/brain/query-patterns.md` Pattern 11 (causal_framework_select) for richer extraction context. Never gate extraction on Brain availability -- extraction works at Tier 0 without Brain.
+
+### Extraction Instructions
+
+Read the artifact content. Identify EXPLICIT causal statements in the text.
+
+Look for these patterns:
+- "X causes Y" / "X leads to Y" / "X results in Y"
+- "Because of X, Y happens"
+- "X enables Y" / "X prevents Y"
+- "When X, then Y (because Z)"
+- "Due to X, Y occurs"
+- "X drives Y" / "X undermines Y"
+
+For each causal statement found, extract:
+
+1. **Cause** -- what produces the effect (quote or paraphrase from text, max 200 chars)
+2. **Mechanism** -- HOW the cause produces the effect (must be specific, not generic -- "through market forces" is NOT acceptable)
+3. **Effect** -- what happens as a result (quote or paraphrase, max 200 chars)
+4. **Extraction method** -- classify how the claim was identified:
+   - `observed` (confidence 0.7): backed by data, measurements, or empirical evidence in the text
+   - `asserted` (confidence 0.5): author states the causal link directly without empirical backing
+   - `inferred` (confidence 0.3): you deduced the causal link from context; mechanism is not explicit in text
+5. **Domain** -- classify based on artifact section and content:
+   - `materials` -- physical materials, manufacturing, chemistry, biology
+   - `business` -- business model, operations, revenue, strategy
+   - `competitive` -- market competition, positioning, differentiation
+   - `financial` -- funding, costs, margins, valuation
+   - `team` -- people, skills, culture, hiring
+   - `legal` -- regulation, IP, compliance, policy
+   - `general` -- does not fit a specific domain
+6. **Falsifiable prediction** -- what testable prediction would disprove this claim? Every claim MUST have one.
+
+**Rules:**
+- **Maximum 5 claims per artifact.** Be selective, not exhaustive. Choose the most significant causal relationships.
+- **Correlation is NOT causation.** "After X, Y happened" is NOT a causal claim unless the mechanism is stated or strongly implied.
+- **Three Gaps enforcement:** Every claim MUST have a non-empty mechanism AND a falsifiable prediction. If the mechanism is unclear from the text, mark as `inferred` with confidence 0.3 and state your best understanding of the mechanism. If you cannot articulate any mechanism at all, do not include the claim.
+- **No generic mechanisms.** "It causes it" or "through various factors" is not a mechanism. Be specific about the causal pathway.
+- If the artifact contains no clear causal statements, say so: "Here's the thing -- this artifact describes conditions but doesn't make explicit causal claims. To build your causal graph, look for statements that explain WHY something happens, not just WHAT happens."
+
+### Presentation
+
+Present extracted claims in a table:
+
+```
+Extracted {N} causal claims from {artifact-path}:
+
+| # | Cause | Mechanism | Effect | Conf | Domain |
+|---|-------|-----------|--------|------|--------|
+| 1 | {cause text} | {mechanism text} | {effect text} | {0.7/0.5/0.3} | {domain} |
+| 2 | ... | ... | ... | ... | ... |
+
+Actions:
+- Accept all: write all {N} claims to your knowledge graph
+- Accept #1,#3: write specific claims (comma-separated numbers)
+- Edit #2: modify a claim before accepting
+- Reject #2: remove from list (provide reason to improve future extraction)
+- Reject all: discard extraction
+```
+
+If fewer than 5 claims found, that is fine -- quality over quantity. Do not pad to fill the quota.
+
+### Confirmation Flow
+
+Wait for the user's response. Do NOT write anything until the user explicitly confirms.
+
+**Accept all:** Proceed to Write and Bridge for all claims.
+
+**Accept specific (e.g., "Accept #1,#3"):** Write only the specified claims. Move others to rejected array without a rejection_reason (they were simply not selected).
+
+**Edit #N:** Let the user modify the claim's fields. Re-present the updated table row for confirmation before proceeding.
+
+**Reject #N (with reason):** Move claim to rejected array. If the user provides a reason, capture it as `rejection_reason`. Rejection reasons are valuable data -- they teach the system what NOT to extract (Decision 13: rejection is data).
+
+**Reject all:** Discard all claims. Ask if the user wants to try a different artifact.
+
+### Write and Bridge
+
+After confirmation, write the results:
+
+1. **Generate claim IDs:** For each accepted claim, generate an ID: `causal-` prefix + 8 hex characters. Use the Write tool to create IDs inline (e.g., `causal-a1b2c3d4`).
+
+2. **Write .causal-extract.json** to the room root directory using the Write tool:
+
+```json
+{
+  "source_artifact": "{artifact-path-relative-to-room}",
+  "extracted_at": "{YYYY-MM-DD}",
+  "claims": [
+    {
+      "id": "causal-{8hex}",
+      "cause": "{cause text}",
+      "mechanism": "{mechanism text}",
+      "effect": "{effect text}",
+      "confidence": 0.7,
+      "extraction_method": "observed",
+      "domain": "materials",
+      "falsifiable_prediction": "{prediction text}",
+      "evidence": ["{source_artifact}"]
+    }
+  ],
+  "rejected": [
+    {
+      "cause": "{rejected cause}",
+      "effect": "{rejected effect}",
+      "rejection_reason": "{user's reason or empty string}"
+    }
+  ]
+}
+```
+
+3. **Call the bridge script** to write confirmed claims to the room graph:
+
+```bash
+node "${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/scripts/causal-to-graph.cjs" "room/"
+```
+
+4. **Handle bridge errors:**
+   - If the room graph is not available (no `.mindrian/room.db`): Report that JSON was saved and claims can be written to the graph later when the graph is initialized. The `.causal-extract.json` file is the durable record.
+   - If the bridge script reports skipped claims (missing mechanism or prediction): Report which claims were skipped and why.
+
+### Post-Write Report
+
+After the bridge completes, report:
+
+```
+{checkmark} {N} causal claims written to knowledge graph from {artifact-path}
+{N rejected} claims noted (not written to graph)
+
+The claims are now queryable via /mos:graph -- try "What causal chains exist in {section}?"
+```
+
+If any claims were skipped by the bridge due to validation failures, note them:
+
+```
+{warning} {N} claims skipped by bridge (missing mechanism or prediction)
+  These claims were saved in .causal-extract.json but not written to the graph.
+  Edit the JSON and re-run the bridge, or use /mos:causal extract again on the artifact.
+```
+
+## Error Handling
+
+**No artifact path provided:**
+Ask the user for a path. Suggest using Glob to find artifacts: "Which artifact? Give me a path relative to room/ -- or I can list what's there."
+
+**Artifact not found:**
+```
+x Artifact not found: {path}
+  Why: No file at room/{path}
+  Fix: Check the path -- use /mos:graph to see what's in your room
+```
+
+**No causal statements found:**
+Explain what to look for rather than returning an empty table. Help the user understand what makes a causal claim vs. a descriptive statement.
+
+**Room graph unavailable:**
+```
+{warning} Knowledge graph not initialized
+  JSON saved to room/.causal-extract.json (claims preserved)
+  Fix: Run /mos:setup graph then re-run the bridge with: node "${MINDRIAN_OS_ROOT:-${CLAUDE_PLUGIN_ROOT:?MindrianOS install root not found. Set MINDRIAN_OS_ROOT (see lib/core/active-plugin-root.cjs) or run from Claude Code.}}/scripts/causal-to-graph.cjs" room/
+```
+
+## When Complete
+
+After extraction, suggest next steps based on what was found:
+- If claims span multiple domains: "These claims cross {N} domains. Run /mos:graph to see how they connect to your existing knowledge."
+- If confidence is mostly inferred: "Most claims are inferred -- look for data to strengthen these. Run /mos:causal extract on artifacts with empirical evidence."
+- If fewer than 3 claims: "This artifact had few causal claims. Try extracting from artifacts in sections where the venture's theory of change is strongest."
+
+Do NOT suggest trace or predict -- those ship in v1.7.0.