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.

The conversation is the whole interface — and it adapts to what you’re actually doing. Every message you send is classified by intent before Larry responds, and that classification sets the conversation mode. You don’t switch modes manually. Larry reads what you’re describing, shifts the mode to match, and changes how he engages accordingly. A casual question gets a different Larry than an active room-building session, which gets a different Larry than a framework running to completion.

​
The five modes

ModeTriggerWhat Larry Does
JUST_TALKExploratory, open-ended input — you’re thinking out loudAsks clarifying questions, reframes the problem, doesn’t rush to file anything
EXPLORE_CAPTUREYou describe something specific enough to be worth filingTakes notes, maps insights to room sections, builds context without interrupting flow
BUILD_ROOMActive construction — commands are firing, sections are populatingRuns commands, files artifacts, surfaces what just changed and what contradicts what
METHODOLOGYA named framework is running — a sequence of steps is in progressSequences the methodology steps, surfaces the chain, holds you to the method’s logic
DECISION_GATEA finding needs your decision — approve, reject, or deferPresents a structured card with the finding, the reasoning, and your three options

​
How mode shifts work

A UserPromptSubmit hook fires on every message. Before Larry responds, the hook classifies the intent of your input and sets the current mode. This is deterministic routing — the mode is a function of what you said and what the room currently knows, not a guess. Mode shifts are not announced. Larry simply engages differently. A turn that starts in JUST_TALK can shift to EXPLORE_CAPTURE mid-session when you describe something concrete enough to file. A methodology completing its last step shifts to DECISION_GATE when it produces a finding that needs your judgment.

​
Three ways to enter a session

MindrianOS recognizes three entry modes, and you can use any of them at any point:
Just talk. No command needed. Describe what you’re doing, what you’re stuck on, or what you learned. Larry classifies your intent and routes to the right method.
“I interviewed three customers last week and I’m not sure what to make of what they told me.”
Larry reads this as EXPLORE_CAPTURE heading toward /mos:file-meeting or discovery work, and responds accordingly.

​
The F-shape selector

The F-shape selector is Larry’s primary way of proposing what to do next. When the methodology graph suggests a framework chain, or when the room state indicates a natural next step, Larry renders a ranked card: 
Choose next reach:
  ▶  /mos:bono         — run six-hats debate on your sub-domains   [RECOMMENDED]
     /mos:analyze-needs — map who has this problem and how badly
     /mos:research       — pull evidence from web + Brain methodology

  [accept]  [defer]  [reject]  [none fit]
The top recommendation carries a RECOMMENDED marker only when the Brain’s confidence is 0.70 or higher. Below that threshold, the marker is absent — the card surfaces options without overclaiming. Selecting “none fit” records a ranker-miss with your intent captured, so Larry can propose something else without losing what you were trying to accomplish.

​
The autonomy gate

For chains of commands that can safely run unattended, use: 
/mos:act --chain
Larry computes the framework chain from your room state, identifies which steps are marked autonomous_safe, and runs the safe prefix unattended. The chain stops at the first step that requires a human judgment call — a Decision Gate, an approval, a choice between directions — and hands control back to you. The gate is structural, not a preference setting. Steps that need you will always need you. The autonomy gate makes the unattended work explicit rather than hiding it.

​
Mode-aware session behavior

Larry adapts not just to what you say but to the shape of the session itself. Four session-level signals shift how Larry engages:
If Larry has redirected you twice in a session — steered you away from a direction with insufficient evidence, or caught a contradiction — he enters follow-mode. Questions become more pointed. He’s tracking a thread.
If the session goes quiet for more than five minutes, Larry surfaces a gap the room currently has — an unanswered question, a section that’s empty despite having upstream input, or a contradiction that hasn’t been resolved.
After eight or more turns in a session, Larry checks whether the conversation has reached saturation — whether you’re covering new ground or circling. If the density of new insight is dropping, he may suggest filing what you have and shifting to a different method.
When you return to a room after time away, Larry opens with a brief: what the room currently knows, what changed since your last session, what the open questions are, and where the most important gaps sit. No re-explaining context you already set.

​
Decision Gate mode

DECISION_GATE is the mode where the work stops until you weigh in. It appears when:
  • A methodology surfaces a finding that needs approval before it files
  • The opportunity harvest produces a qualified candidate
  • A research result contradicts something already in the room
  • The room detects a gap that could resolve in multiple ways
At a Decision Gate, you have exactly three options: APPROVE, REJECT (with a reason), or DEFER. All three are recorded as typed graph data. A rejection reason becomes part of the room — it informs the next scan, the next recommendation, and the next time Larry encounters something similar. This is not a formality. Canon Part 3 of the MindrianOS architecture specifies that material decisions — anything that changes the room — require a human at the gate. Larry proposes; you decide.

Build docs developers (and LLMs) love

Get started for freeTalk to us