Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/jsagir/mindrian-os-plugin/llms.txt

Use this file to discover all available pages before exploring further.

When you want to ask the knowledge graph a question without writing SQL, /mos:graph translates plain English into the right traversal. Larry walks the edges so you do not have to. The graph builds automatically as you use your room — every filing from every command wires typed nodes and edges into the local SQLite database. No manual graph construction needed; the intelligence accumulates as you work.

​
Usage

/mos:graph "<natural language question>"
/mos:graph --derive
ArgumentDescription
"<question>"A plain-English question about your room’s knowledge graph
--deriveRun the HEAL-FIRST backfill — wires an existing room and its sub-rooms in one pass

​
The Room Graph

The room graph is a SQLite database at room/.mindrian/room.db. It is the local, private accumulation of everything Larry has filed into your room — artifacts, sections, claims, evidence, and the typed relationships between them.
All graph reads and writes are strictly local. Zero room data reaches the Brain. The Brain (Neo4j + Pinecone) holds methodology intelligence — your room data never crosses that boundary.

​
Typed Edge Vocabulary

Edge typeMeaning
INFORMSOne node contributes evidence or reasoning to another
CONTRADICTSTwo nodes make conflicting claims
CONVERGESMultiple nodes independently support the same conclusion
INVALIDATESOne node renders another’s claim no longer valid
ENABLESOne node is a prerequisite for or unlocks another
REJECTED_BECAUSEA rejected finding or candidate, with the captured reason scalar
SUPERSEDESA higher-tier source replaces a weaker prior claim
BELONGS_TOAn artifact or claim belongs to a section

​
What Happens

1

Pre-flight check

Larry checks that room/.mindrian/room.db exists. If the graph is empty or missing, it explains how to start building it: “The graph builds as you file content — meetings, documents, methodology sessions.”
2

Graph stats on first invocation

On first use, Larry surfaces a summary: total node count, edge count, and section coverage. This orients you before you start querying.
3

Natural language to SQL translation

Your question is translated into a SQL query against the graph schema. Common patterns:
  • “What connects X to Y?” → edges between sections X and Y
  • “Where are the contradictions?” → edges where type = 'CONTRADICTS'
  • “Which sections are most connected?” → edge count per section, ordered by frequency
  • “What are the gaps?” → sections with few or no outbound edges
4

Results presented as room cards

Results are never raw query output. Each match surfaces as a card showing the node title, section, relationship type, and a plain-English explanation of the connection in your venture context.
5

Next-move suggestion

After results, Larry suggests the most useful next action — a follow-up query, a related command, or an invitation to derive new typed edges via /mos:graph --derive.

​
Example Queries

# Find the weakest assumption in your financial model
/mos:graph "what is the weakest assumption in my financial model?"

# Surface all contradictions in the market section
/mos:graph "show me all CONTRADICTS edges in the market section"

# Find what connects two sections
/mos:graph "what connects problem-definition to market-analysis?"

# Find coverage gaps
/mos:graph "which sections have the fewest connections?"

# Find topics that appear across multiple sections
/mos:graph "what topics appear in 3 or more sections?"

​
The --derive Backfill

/mos:graph --derive wires an existing room and all its sub-rooms in one pass. It is the universal net for surfaces where automatic graph hooks do not fire (Desktop / Cowork). The backfill runs in a strict HEAL-FIRST sequence:
1

Step 0 — GDH-08 self-heal

Detects any artifact-bearing folder that lacks its own .room-root sentinel. For each one found, Larry surfaces a Decision Gate card per folder — APPROVE, REJECT (with reason), or DEFER. Only approved folders are healed. A rejected folder is surfaced but left untouched.
2

Step 1 — resolve

Resolves the now-sentineled room by resolveRoomRoot.
3

Step 2 — rebuild transitively

Rebuilds the graph root-files-aware, non-.md-aware (.docx / .html via the local extractor), and sub-room recursive at arbitrary depth.
4

Step 3 — derive per room and sub-room

Calls runDerivation once per room and per healed sub-room. The default deriver is score-based (not regex keyword matching) — it emits CONVERGES and INFORMS edges from real semantic similarity. All derived edges land as proposed, pending your confirmation at the gate.
5

Step 4 — report the delta

Reports typed-edge count before and after: “Wired [parent] and [K] sub-rooms. Typed edges: [before] → [after]. [N] new relationships are PROPOSED — review and confirm at the gate.”
The --derive backfill is idempotent — re-running it is a no-op. An already-healed room keeps its .room-root sentinel, and the deriver’s pre-propose guard never mints duplicate proposed nodes or downgrades confirmed ones.

​
With Brain: Methodology-Level Queries

When Brain is connected, you can also run Cypher queries directly against the methodology graph via /mos:graph’s Brain integration — surfacing framework chains, prerequisite relationships, and methodology-level connections alongside your local room data. Without Brain, queries run on the local SQLite graph only. All room-level analysis — contradictions, gaps, cross-section connections — works fully in Tier 0.

/mos:research

Wire new EvidenceClaim nodes and typed edges into the graph via web research.

/mos:find-connections

Use Brain semantic search to surface non-obvious cross-domain patterns in the graph.

/mos:opportunities

Browse the Opportunity Bank, which also lives as typed nodes in the room graph.

/mos:qualify-opportunity

Review qualification history and REJECTED_BECAUSE edges filed to the graph.

Build docs developers (and LLMs) love

Get started for freeTalk to us