--- name: mva-pipeline description: Auto-activates when UserPromptSubmit detection classifies the user's prompt as a venture sentence; relays the 30-second MVA brief in Larry's voice license: BSL-1.1. See LICENSE for complete terms (Business Source License 1.1, Change Date 2030-04-16 to Apache License 2.0). auto-activate: state-file state-file: ~/.mindrian/mva/.json state-condition: pending && !running interactive_first_reward: instant_brief canon_parts: [Part 2, Part 8, Part 10] # --- Phase 143.3 connector frontmatter --- connector: connects_to_spine: true sensor_triggers: [] reach_id: context_block sub_mode: mva-pipeline posture: push_forward hierarchy_rank: 12 filing: memory_event_only plan_gated: false web_scope: null surface: F.1 hitl_shape: "F.1" hitl_why: "The 30-second brief closes with a numbered option-or-free-text choice (1/2/3/free-text), the canonical F.1 Next Move shape." layer: "loop" layer_why: "Detects a venture sentence and relays the pre-computed 30-second MVA brief through one F.1 next-move choice; a single bounded cycle to a stopping condition, not new methodology or scaffolding." --- At this skill'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 skill'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. # The 30-second MVA skill ## When this activates When Plan 118-00's UserPromptSubmit detection writes a pending state with `pipeline_status: 'pending'` (or `hebrew_refusal: true`), this skill fires on the NEXT model turn. The skill is the bridge between Plan 118-00's classifier and Plan 118-03's orchestrator: it makes Larry the GUIDED narrator of the brief without baking the orchestrator into Claude Code's hook protocol. ## What to do 1. Run `/mos:mva-brief` (or invoke `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/mva-run.cjs"` via Bash directly). 2. Relay the stdout to the user VERBATIM, in your normal Larry voice (no extra framing). The renderer already speaks in Larry's GUIDED voice; double-voicing breaks the pedagogical contract. 3. Wait for the user's option selection (1, 2, 3, or free-text). 4. Route per the footer behavior: - 1 -> stay in JUST_TALK; the brief stays visible; user can ask follow-ups about any cell ("tell me more about the Honeydue case", "what was the Gottman analogy?", etc.) - 2 -> invoke /mos:new-project (Phase 119 wrapper; in v1.13.0 this surfaces a stub message that the full room-build flow lands in beta.18) - 3 -> invoke /mos:challenge-assumptions against the brief ## What NOT to do (per feedback_larry_pedagogical_guided_first.md) - Do NOT add commentary or "I noticed..." preamble before the rendered output. The renderer already opens with "Scanning for precedents..." in Larry's GUIDED voice. - Do NOT interpret findings autonomously ("This means you should..."). The rendered output IS GUIDED -- it asks the user to think; the renderer surfaces what's in the graph and lets the navigator chew on it. - Do NOT skip the 3-option footer. Even if all agents failed, the sharp-question fallback substitutes for the footer (per binding decision B7); the renderer handles this -- you just relay verbatim. - Do NOT pre-pick an option. The user picks. That's the decision gate. ## Canon parts implemented - Part 2 (team around navigator -- 6 agents as a parallel team, with the skill being the surface Larry uses to relay the team's output) - Part 8 (boundary -- the agents send ONLY generic handles to Brain and Tavily; the renderer + skill never re-introduce user content) - Part 10 sub-claim 3 (room as receipt -- the brief IS the reward delivered BEFORE the user is asked to invest in setting up a room; option 2 is the ask-for-investment that comes AFTER the reward, per Hooked sequencing) ## State file contract Read from `~/.mindrian/mva/.json` (the Plan 118-00 wire): ``` { "sentence_sha256": "<64-hex>", // NEVER the raw sentence "classified_at": , "classifier_source": "heuristic" | "heuristic_fallback" | "haiku-4-5", "classifier_confidence": "high" | "medium" | "low", "locale": "en" | "he", "hebrew_refusal": true, // optional, set on LD1 short-circuit "pipeline_status": "pending" | "running" | "complete" } ``` On `hebrew_refusal:true`, the orchestrator renders the bilingual refusal block and DOES NOT fire the dispatcher (per LD1 in 118-CONTEXT.md). ## Routing the 3-option footer (after the brief renders) Once `/mos:mva-brief` (or the orchestrator) has rendered the brief to the user, the user's next message is most likely an option selection. Recognition rule: - User types exactly `1`, `2`, or `3` -> invoke `/mos:mva-option ` (no sha8 argument needed; the command auto-resolves via `resolveCurrentSha8()`) - User types `/mos:mva-option N` explicitly -> invoke `/mos:mva-option ` (sha8 still optional; auto-resolved when omitted) - User types anything else -> handle as a normal conversation turn (do NOT route through `mva-option`) The sha8 argument is OPTIONAL because the router auto-discovers the most recent brief via `resolveCurrentSha8()` -> `~/.mindrian/mva/state.json` (the manifest written by Plan 118-03's orchestrator after `mva_brief_rendered`). ### Per-option behavior Option 1 -- "Just tell me what's new" (stay in tell-me mode): - Acknowledgment: "Keeping the brief visible. Ask me anything about what you just saw." - Operator: transitions to `JUST_TALK` - Brief stays in scrollback; follow-up questions about any of the 6 cells are welcome Option 2 -- "Build a room around this" (invest, deferred): - Show the stub message verbatim from `STUB_MESSAGE_119`: "Building a room around this is the next layer; shipping in beta.18 (Phase 119). For now, press option 1 to keep this brief visible, or option 3 to go deeper." - Operator: no transition (option 2 is stubbed for v1.13.0) - In v1.13.0-beta.18 (Phase 119), this routes to `/mos:new-project --from-brief ` Option 3 -- "Challenge me -- Devil's Advocate" (go deeper): - Bridge text: "Going deeper. Pulling the brief into a Devil's Advocate pass." - Operator: transitions to `METHODOLOGY` - Invoke: `/mos:challenge-assumptions --from-brief ` ### Edge cases - Brief data expired (side-file missing): tell the user the brief expired and offer to re-run by typing their venture sentence again. - Brief is still rendering (`mva_brief_rendered` event not yet emitted): hold the option, tell the user "Brief is still rendering -- options will activate when it completes". - No `state.json` exists (fresh install or Hebrew refusal path): tell the user "No recent brief found. Type your venture sentence to fire the pipeline." - Invalid option (`4`, `99`, etc.): treat as free-text and route normally. ### Do NOT - Do NOT autonomously pick an option for the user. - Do NOT pre-summarize what option 2 "would do" -- the stub message says it. - Do NOT add em-dashes to any rendered option text -- use `--` only. - Do NOT invoke `/mos:new-project` for option 2 in v1.13.0 -- that wiring lands in Phase 119 (beta.18).