MindrianOS Release Notes and Full Version Changelog

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.

​

v1.15.3-beta.28 — 2026-07-18

​

Added

• Eureka Seam 2 fix (statement-metadata gap). Every entity-entity Opportunity Statement was rendering the literal placeholder “unknown x unknown approach to a unknown x unknown cross-domain bridge” instead of a real mechanism. The entity-node branch in lib/core/eureka/room-native-substrate.cjs now inherits section from its DESCRIBES edge to its source memory artifact, and scripts/eureka-portfolio-report.cjs uses relation-edge-typed bridge labels (competes-with/uses-component/supplies-to) instead of a generic phrase. Result: 22/25 → 0/25 “unknown x unknown” statements on both proving rooms.
• Eureka Seam 3 fix (candidate-generation gap). Section container nodes — the room’s own top-level folder nodes — were admitted as pairing candidates, flooding ranked-list slots with domain-same pairs. An additive either-endpoint Section exclusion at the generation layer now excludes these. 1,575 degenerate pairs (9.8% of the candidate set) excluded on iia-deeptech-centers; proven a true no-op on rooms with no Section nodes.
• Warm-cache MCP path for Eureka. eureka-run, eureka-status, and eureka-report are now available on the intelligence MCP router tool. They call the same governed dispatcher (scripts/eureka-command.cjs) in-process instead of spawning a child process per call, so the embedding encoder cache stays warm across scans on the Phase 198 resident daemon. Measured: same-process call 2 is 0ms vs. call 1’s 179ms.
• Phase 230: MindrianOS Skill Fleet Optimization harness. A harness for testing whether all 124 skills trigger correctly and stay quiet when they should not. Two workstreams: WS1 (trigger accuracy across all 124 skills) and WS2 (code quality for the ~59 script/workflow-backed skills). Smoke-tested end to end on a 13-skill calibration set. The full 124-skill fleet run is explicitly deferred behind a future opt-in — this release ships the harness only.

​

Fixed

• check-card-fire.cjs no longer force-fires the Decision-Gate card on plain prose with no actual gate. Two over-fire mechanisms retired: (1) the backstop’s bare numbered-prose detector had a 7/7 false-positive rate on logged evidence and zero true catches; (2) the primary registry-gated path fired on ran_entries alone due to a session-key TTL bleed. Now requires a confirmed, non-empty gate-subject plus relevance. Verified against all 7 real logged firings (0/7 re-fire).
• Per-session room binding no longer re-prompts every turn after a real bind. The MCP room_bind tool keyed binding state by the Claude session UUID; the CLI hook read it keyed by process.env.CLAUDE_SESSION_ID (unset in that context), silently falling back to a hash that never matched. The hook now reads the session ID from its own stdin payload.
• The reach/navigation dial no longer offers a topically-unrelated room. A structural relevance gate (lib/hmi/reach-relevance-gate.cjs) now suppresses off-topic candidates by token overlap against the live turn before they are offered.

​

v1.15.3-beta.26 — 2026-07-16

​

Added

• Phase 227: Ignite mode-select timing across turns 1–4 (SEED-060). The session-start mode-selection Decision Gate (Just Talk / Explore+Capture / Build a Room) gets a structural, advisory-only backstop: a new doctor.cjs check class detects a silent skip and warns — it never blocks or re-fires the gate itself. skills/larry-personality/SKILL.md now names ignite for the first time and documents the gate’s timing with Fogg B=MAP / TARI reasoning. Mode 3 (Build a Room) now routes through ignite’s Directive/--express path instead of calling /mos:new-project directly.
• Phase 229: HUJI Pitch Feedback Module (in progress, 8/9 plans). MindrianOS’s batch pitch-feedback pipeline: turn each student’s diarized 5-minute pitch transcript into one Minto-structured formative feedback artifact, batch-orchestrated across 200+ submissions at a $4–5/unit cost ceiling, local-only scoring (Brain read-only, generic handles per Canon Part 8). Ships the evidence/feedback schemas, test-inventory harness, deterministic code checks, PWS_grading recipe, Stage A intake adapter, LLM judge spawner, and batch orchestrator. The mandatory demo run and human verdict checkpoint have not yet shipped.

​

v1.15.3-beta.24 — 2026-07-16

​

v1.15.3-beta.22 — 2026-07-16

​

v1.15.3-beta.20 — 2026-07-16

​

Added

• Phase 222 (reach-ranking-unification). The three surfaces that suggest a next move (/mos:suggest-next, the reach-candidates list, the auto-fire engine) now always agree on the top pick instead of occasionally diverging, and ranking improves over the room’s own accept/reject history.
• Phase 223 (JTBD-driven intelligence pipeline). /mos:bono is now an 8-phase governed research debate with Six-Thinking-Hats-style scrutiny (the Black hat must disconfirm first, the White hat must cite-or-retract) and three explicit navigator approval pauses. New /mos:intel-pipeline command runs a staged research pipeline oriented on the room’s active JTBD (calibrate → decompose → fan out → compute → synthesize → close).
• Phase 224 (graph-derivation-harness, SEED-034). Every markdown write to a room now enqueues and background-derives typed graph edges. No manual “derive” step is required.
• Phase 225 (per-session-room-binding, SEED-039). Session-to-room binding now correctly supports multi-room binds.
• Phase 226 (eureka-reasoning-mode-fallback, SEED-058). When the local embedding encoder is unavailable, /mos:eureka degrades to a labeled REASONING MODE result (a real short ranked list with an honest caveat naming the degrade cause) instead of a hard pairs_scored: 0 dead end.
• Eureka entity extraction two-tier WHAT-vs-WHY classifier. A free, fully local embedding pass now resolves the confident majority of candidates at zero API spend. The existing LLM classifier is demoted to an escalation-only path for ambiguous residuals. Measured on a real room: 61.1% of candidates resolve locally and correctly, 14.3% fewer LLM calls. Tunable via MINDRIAN_WHATWHY_MARGIN (default 0.10).

​

Fixed

• Eureka ranked top-25 no longer refills with scaffold pairs when real entities are thin. Scaffold-vs-scaffold pairs are excluded from the candidate set by construction; the exclusion count is reported in provenance.
• Eureka entity-extraction pre-step no longer fails silently — failures now surface as an extraction_error field in eureka status.json plus a stderr line.

​

v1.15.3-beta.18 — 2026-07-13

​

v1.15.3-beta.16 — 2026-07-13

​

Added

• Opportunity follow-through lifecycle. Every opportunity flows through the Harvest Formula (candidate → qualified → explored → promoted | parked | retired) as a real graph node with append-only stage history.
• Eureka statements bank as proposed opportunity nodes. Statements the Grounding Guard passes bank as opportunity nodes with DERIVED_FROM evidence edges. Tunable via MINDRIAN_OPPORTUNITY_BANK_PREDICATE (critic | critic+tail | all).
• Harvest sensor (SENS-14). A producer on the insight-sensor rail harvests candidates from five lanes (eureka proposals, bridges, contradictions, whitespace, meeting filings), classifies each through the Gibson Four-Lens, and scores them with HarvestIndex_v1.
• Qualification Decision Gate (/mos:qualify-opportunity). Five verbs: Qualify+file, Park, Retire, Explore, Skip. Nothing qualifies without your explicit verb.
• /mos:explore-opportunity. One explicit action turns a qualified opportunity into deep research — a Minto-shaped opportunity artifact filed into opportunity-bank/ plus a research corpus artifact in research/, with typed evidence edges.
• Web ingestion agent. Paste a URL or run /mos:research <url> — after you approve at the card, the page is fetched, filed as a cited research artifact in research/, and its entities land in the room graph. Nothing is fetched or filed without your explicit verb.
• Pasted-URL sensor (SENS-15). A bare URL in your turn offers an ingest card contextually. Never auto-files.
• Content-hash idempotency + SUPERSEDES versioning. Re-ingesting an unchanged page is an honest no-op; a changed page files a new version linked to the prior one.

​

v1.15.3-beta.14 — 2026-07-12

​

Fixed

• check-card-fire.cjs no longer force-fires on benign numbered lists. A bare numbered-prose list (1. / 2.) now triggers the backstop only when a choice-framing cue (?, which, would you like, pick, choose, select, or type 1) appears inside the matched span or within the preceding ~150 characters. Genuine bracket-notation, type 1, 2, or 3 literal, and multiline bracket-box alternatives remain unconditional.
• The Stop-hook reason slug no longer surfaces as a fake “Stop hook error” message. The reason field is now a calm, human-safe phrase on both the intercept and degrade branches. The internal slug is preserved in the new local diagnostic log only.
• New local-only intercept diagnostic log (~/.mindrian/card-fire-intercepts.log). Append-only JSONL capturing timestamp, session ID, original reason slug, gate signature, and truncated output text. TTL-pruned on every write. Local disk only — never the Brain, never a network wire.
• gate-is-simple-binary exemption no longer swallows genuine two-option forced-choice forks. The exemption now requires labels to be YES/NO-shaped (via isYesNoShapedGate), not merely 2-in-number, so a real two-way strategic fork fires the card like a 3+-way fork would.
• MCP room_state reads re-resolve the active room per call. The status, analyze, compute-state, get-state, and suggest-next branches previously read a boot-time closure roomDir, making mid-session room switches invisible and causing status to report “No room initialized” against rooms with real content.

​

v1.15.0 — 2026-07-02

​

Changed

• Phase 210: Revert persona-enforcement over-reach. Five phases in the v1.15 window turned voice/gate judgment calls into hard-fail mechanical checks. Each was softened to advisory or relevance-gated behavior while underlying capabilities stayed intact: shape-declaration gate is advisory (exits 0, still detects and enumerates violations); voice-glyph mapping is a default, not a lock; APO voice-contract scores rather than vetoes; elevation quorum suggests rather than forces; force-fire is relevance-gated. The Phase 194 session-room-binding gate, Phase 196 Brain-egress PreToolUse guard, and Phase 200 semantic-floor gate are byte-untouched.

​

Added

• Phase 190: Shape-F Declaration Mandate (Canon v1.23, Part 11 R16). Every invocable surface now has a declared HITL shape, enforced at commit, release, and doctor --acceptance.
• Phase 192: HITL Selector Completion + Posture Dial + live statusline. Menu live-selectors across suggest-next/rooms/onboard, the F7max dial preview and modifier pane, new /mos:stance 4-pole posture dial, and a statusline that updates per turn.
• Phase 203: Synthetic-Expert-as-Skill with a two-surface Plurai construction + behavioral fidelity gate.
• Phase 204: Ignite Room-Chooser + persona-differentiated entry, the first consumer of the Shape-F selector front door.
• Phase 187.2: CLAUDE.md de-bloat. Fully-expanded per-session load reduced from ~66k tokens to ~3,899 tokens (under the 4k gate), proven durable via idempotent regeneration.

​

Versioning