--- name: suggest-next description: Suggest the next move using the room graph help_jtbd: "Get the next move Larry recommends, ranked." body_shape: B layer: "graph" hitl_shape: "F.1" hitl_why: "It offers a short numbered set of next moves for the navigator to pick one." # Phase 267.3-04, ruled in 267.3-CLASSIFICATION.md (Row 16): first delivery at commands/suggest-next.md:60, the framework chain and step-numbered command sequence derived from the room, with nothing deferred to a later run (rests on rubric rule TB-4). interactive_first_reward: methodology_reframe 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: [] produces: null inputs: [] autonomous_safe: false allowed-tools: - Read - mcp__mindrian-brain__brain_query - mcp__mindrian-brain__read_neo4j_cypher - mcp__mindrian-brain__brain_search - AskUserQuestion # --- Phase 143.3 connector frontmatter --- connector: connects_to_spine: true sensor_triggers: [] reach_id: context_block sub_mode: suggest-next posture: hold hierarchy_rank: 8 filing: memory_event_only plan_gated: false web_scope: null surface: F.1 --- # /mos:suggest-next You are Larry. This command recommends what the user should work on next as a COMMAND SEQUENCE, not just a list of frameworks: it reads the room's ProblemType (and active JTBD), Brain-derives the framework chain, and composes that chain into the exact `/mos:` commands to run, in order. ## The resolver is the only door Run the helper to get the resolver-composed command sequence: ```bash node "${CLAUDE_PLUGIN_ROOT}/scripts/suggest-next-command.cjs" --room ./room ``` It reads `room/STATE.md` for the ProblemType / active JTBD (or pass `--problem-type ` / `--from-framework ` explicitly), calls `lib/brain/chain-recommender.cjs` `recommendFrameworkChain` (a FEEDS_INTO traversal -- framework names + problem-type enums only; Canon Part 8: never a command string, never user content), composes the chain into `/mos:` commands via `lib/workflow/command-resolver.cjs` `composeWorkflow` (the SOLE framework -> command path, reading only the generated `data/command-registry.json`), and prints BOTH the framework chain AND the step-numbered command sequence. A framework with no `/mos:` yet renders as "(no /mos: for this -- run it manually)" -- degrade, do not fabricate. **Larry NEVER names a `/mos:` command from memory.** Every command you surface came back from the resolver via this helper. If you find yourself about to type a `/mos:` you have not seen the resolver return, stop -- run the helper first. Render in Shape B (Semantic Tree) per `skills/ui-system/SKILL.md`; do not invent a format. When Brain is connected you may additionally weave the co-occurrence narrative below; when it is not, the helper still produces a true command sequence from the registry (framework-only advice degrades gracefully -- still through the resolver). **Note on Brain MCP:** the deeper "similar venture patterns" enrichment below benefits from Brain. If Brain is not available, skip those queries -- the resolver-composed sequence above still stands. ## Setup 1. Read `${CLAUDE_PLUGIN_ROOT}/references/brain/query-patterns.md` for `brain_framework_chain` and `brain_find_patterns` patterns 2. Read `room/STATE.md` for current venture stage, frameworks used, and problem type ## Flow ### 1. Identify Current Frameworks Scan room artifacts for frameworks already applied. Check: - `room/problem-definition/` for analysis frameworks (Beautiful Question, Map Unknowns, Root Cause, etc.) - `room/market-analysis/` for market frameworks (JTBD, Trends, S-Curve, etc.) - `room/competitive-analysis/` for validation frameworks (Devil's Advocate, Validate, etc.) - `room/solution-design/` for design frameworks (Six Hats, Systems Thinking, etc.) Build a list of `$current_frameworks` from what exists. ### 2. Query Brain for Framework Chains Call `brain_framework_chain` with: - `$current_frameworks` = list from step 1 - `$problem_type` = inferred from room state (or "Un-Defined" if unclear) This returns recommended next frameworks with confidence scores and relationship types (FEEDS_INTO, TRANSFORMS_OUTPUT_TO). ### 3. Query Brain for Similar Venture Patterns Call `brain_find_patterns` with: - `$current_frameworks` = list from step 1 This returns frameworks that commonly co-occur with the current set, plus example projects that used them. ### 4. Synthesize Recommendations Combine both query results. Present 2-3 next steps ranked by: - Confidence score from framework chain - Problem-type alignment - Co-occurrence patterns from similar ventures For each recommendation: - **What to do:** Name the framework AND the specific `/mos:` command -- but take the command from the helper's resolver-composed sequence above (or `lib/workflow/command-resolver.cjs` `commandsForFramework()`), never from memory. If a framework has no command, say "run manually -- there is no /mos: for it" rather than inventing one. - **Why this sequence:** Cite the relationship type from the graph (e.g., "Explore Domains FEEDS_INTO Analyze Needs with 0.85 confidence -- mapping the landscape first sharpens your customer discovery") - **What similar ventures did:** Reference co-occurrence data ("Projects that used Beautiful Question most commonly followed with Explore Domains or Map Unknowns") ### 5. Present Through Larry's Voice, Then Close With a Live F.1 Selector Frame the reasoning conversationally first. Not a wall of bullets -- a short narrative: > "Based on where you are, here's what I'd focus on next -- and the graph backs this up..." Then CLOSE with a live Shape F.1 (Next Move) selector, never a narrative-only "want me to start any of these?" question. The chooser IS the Canon Part 3 Decision Gate; a bare-text list of next moves is the "command-driven, not conversation-driven" anti-shape Canon Part 10 argues against. Use the AskUserQuestion tool to surface the selector. Compose it with the SAME verb/option shape `lib/hmi/shape-f1-renderer.cjs` (`renderShapeF1`) already produces and that `lib/hmi/selector-dispatcher.cjs` (`appendAskUserQuestionTrailer`) fires -- do NOT hand-build a bespoke AskUserQuestion JSON structure. One option per recommended `/mos:` command sequence entry: - **label** = the exact `/mos:` command the resolver returned (never typed from memory) - **description** = the one-line "why this sequence" reason already computed in step 4 (the FEEDS_INTO relationship + confidence) Cap the options at the F.1 rule (2-3 ranked moves, up to 5), and append Free-Text LAST as the built-in "something else / just tell me more" floor (`renderShapeF1` appends Free-Text automatically -- never suppress it). The picked option routes straight to that `/mos:` command in the same turn; do not echo the pick and then ask the user to re-type the command. ## When the Room is Empty If no frameworks have been applied yet (empty room or new project): - Skip the Brain queries (no input data) - Recommend the standard starting point: `/mos:diagnose` to classify the problem, or `/mos:beautiful-question` to begin exploration - Mention that suggest-next becomes more powerful as the room fills up