@mindrian_os/install 1.13.0-beta.28 → 1.13.0-beta.30

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "mos",
   "description": "MindrianOS -- Your AI innovation co-founder. Larry thinks with you through PWS methodology, builds your Data Room as you explore, and chains frameworks intelligently. Install and go.",
-  "version": "1.13.0-beta.28",
+  "version": "1.13.0-beta.30",
   "author": {
     "name": "Jonathan Sagir",
     "url": "https://mindrian.ai"

package/CHANGELOG.md

@@ -1,3 +1,18 @@
+## [1.13.0-beta.30] - 2026-05-23
+
+### Fixed (Engine 1 Act 1 silent-failure class, Phase 127.2 Plan 03 -- FIRST hotfix from external tester evidence)
+- **`scripts/doctor.cjs --check-rs-engine` pre-flight pre-flights Python deps for the Engine 1 Act 1 surface (Finding F1 -- `.planning/debug/resolved/windows-tester-find-bottlenecks-silent-failure-qa-sweep.md`).** ADD-ONLY flag handler (NOT a class flag; owns its own exit-code contract). Probes critical deps reachable from `scripts/rs-engine.py` -- `requests` (transitive via `lib/core/rs_corpus.py`, the actual silent-failure root cause on the Windows tester's machine) + `numpy` -- plus umbrella deps from `requirements-hsi.txt` (`sentence_transformers`, `sklearn`). Resolves python interpreter via `MINDRIAN_PYTHON` env override > `python3` > `python` fallback. On missing critical deps: exit 1 + actionable fix line `Run: pip install -r requirements-hsi.txt --user (use python -m pip if pip is not in PATH)`. JSON variant returns `{ ok, ready, python, probes[], missing, missing_critical, missing_umbrella, fix }`. Defensive: any uncaught probe error surfaces as exit 1 -- the probe never crashes `/mos:doctor`'s other gates. Closes the Windows tester 2026-05-23 silent-failure class for the pre-flight surface.
+- **`lib/agents/reverse-salient-agent.cjs` forwards rs-engine stderr to `result.detail.diagnostic` (Finding F2 -- same RCA).** The `runRsEngine()` catch block now captures `e.stderr` from the failed child python process, takes the LAST 200 chars (preserving the actionable fix line + exception name at the tail), and embeds it in `result.detail.diagnostic`. Backward compatible: when stderr is empty, `detail` stays a plain string (the existing `e.message` slice); when stderr is present, `detail` upgrades to `{ message, diagnostic }`. The existing `ok` / `reason` fields are untouched. Before this fix, every actionable error from rs-engine was silently discarded at the agent layer -- the worst-shape silent failure in a methodology surface.
+- **`commands/find-bottlenecks.md` empty-result UX disambiguates analyzer-down from no-findings (Finding F7 -- same RCA).** New "Empty-result UX" sub-section distinguishes two categorically different cases: (a) analyzer ran with no findings (plausible if room is small or genuinely balanced); (b) analyzer could not start -- detected via `result.detail.diagnostic` or `reason: rs_engine_invocation_failed`. The second path explicitly tells the user to run `/mos:doctor --check-rs-engine` with the pip-install one-liner inline. Before this fix, both cases rendered as "no findings" -- which reads as "your work is clean" regardless of whether the analyzer ran or crashed. The single most dangerous reading in a methodology surface.
+
+### Internal (Phase 127.2 Plan 03)
+- **New test:** `tests/test-127.2-03-rs-engine-silent-failure-fixes.sh` verifies all three findings landed (F1 flag-handler + actionable fix line embedded; F2 diagnostic-field reference + stderr-capture pattern; F7 disambiguation copy present in find-bottlenecks.md or agent file) plus a functional probe asserting `node scripts/doctor.cjs --check-rs-engine --json` returns valid JSON and `--help` documents the new flag. 7/7 PASS on origin/main.
+- **Phase 134 scaffolded as v1.14.0 architectural stub (Finding F6 -- structural answer to install-fragility class).** New `.planning/phases/134-cjs-port-of-python-analyzers-via-xenova-transformers-elimina/134-CONTEXT.md` captures the design vision: replace `scripts/rs-engine.py` + `lib/core/rs_*.py` + `scripts/hsi-*.py` with CJS equivalents using `@xenova/transformers` (ONNX `Xenova/multilingual-e5-large` in-process). Eliminates Python from user-machine surface entirely. Estimate ~3 weeks. Plan 127.2-03's original spec named "Phase 130" but slot 130 was already taken; SDK assigned 134 as next free slot. No PLAN.md (scaffolding only); enters v1.14.0 planning cycle.
+- **RCA closed and moved.** `.planning/debug/windows-tester-find-bottlenecks-silent-failure-qa-sweep.md` -> `.planning/debug/resolved/` with `resolved_by: phase-127.2 Plan 127.2-03 (hotfix; F1+F2+F7 shipped)` + `resolved_disposition: 3-of-4-fixed-in-code; F3 narrative drift deferred to next docs reconciliation; F6 architectural port scaffolded as Phase 134 stub`. Knowledge-base entry appended at `.planning/debug/knowledge-base.md` so `gsd-debugger` surfaces this as a known-pattern hypothesis on future investigations.
+- **Dog-fooding milestone:** the FIRST hotfix shaped from an EXTERNAL tester's transcript (Aryeh's Windows machine, 2026-05-23). Plans 127.2-00 + 127.2-02 were both maintainer-discovered. This one closes a defect a real user hit on a machine the maintainer doesn't own, and ships in the same week the transcript landed -- empirically demonstrating that the dog-fooding loop the QA sweep itself flagged as broken (RS-2 thesis: one-person QA is the lagging subsystem) is no longer one-person.
+
+
+
 ## [1.13.0-beta.28] - 2026-05-23
 
 ### Fixed (post-ship QA-sweep closeout, Phase 127.2 Plan 02)

package/commands/find-bottlenecks.md

@@ -37,6 +37,15 @@ Procedure (CLI / Desktop / Cowork):
    - On DEFER: DEFERRED memory_event records the deferral for Phase 116 unresolved-tension-hook consumption.
 4. If the agent returns `{ ok: false }` OR finds nothing OR is suppressed (tier 0 / JUST_TALK), fall back to the standard Setup + Session Flow below.
 
+### Empty-result UX (Phase 127.2 Plan 03 -- Finding F7)
+
+When the agent returns no findings, you MUST distinguish two cases for the user, because "no findings" reads as "your work is clean" -- the worst possible signal if the analyzer crashed:
+
+- **Analyzer ran successfully and found nothing.** Surface: "No reverse-salient findings were returned -- the analyzer ran across your room and could not identify a lagging component above the threshold. This is plausible if the room is small (less than 5 substantive artifacts) or if the system is genuinely balanced; continue with the Session Flow below to do the framework manually if you want a second pass."
+- **Analyzer could not start (rs-engine failed -- look for `result.detail.diagnostic` in the agent payload, OR `ok: false, reason: rs_engine_invocation_failed`).** Surface: "No reverse-salient findings were returned. If you expected results, the analyzer may not have started -- run `/mos:doctor --check-rs-engine` to verify your Engine 1 Act 1 environment is healthy. Common cause: missing Python deps (`pip install -r requirements-hsi.txt --user`)."
+
+The disambiguation is critical because the agent layer historically swallowed the actionable error message (Windows tester 2026-05-23 silent-failure class). Always surface the `--check-rs-engine` hint on the analyzer-failure path.
+
 Anti-pattern reminder (per docs/AGENTIC-SURFACING-PATTERN.md):
 - Never print findings to console; the F.0 dispatcher IS the surfacing surface.
 - Never query the Brain directly; the agent reads pre-derived BRAIN.md via folder-memory.readQuadruple (LOCAL only, Canon Part 8).

package/lib/agents/reverse-salient-agent.cjs

@@ -189,10 +189,26 @@ function runRsEngine(opts) {
       timeout: 60000,
     });
   } catch (e) {
+    // Phase 127.2-03 Task 1 (Finding F2): forward the child python process's
+    // stderr LAST 200 chars to result.detail.diagnostic so callers can
+    // self-recover from missing-deps / import errors (the Windows tester
+    // 2026-05-23 silent-failure class). Truncate at the START so the tail
+    // (which carries the exception name + actionable fix line printed by
+    // rs_corpus -- e.g. "Run: pip install -r requirements-hsi.txt") is the
+    // half that survives the cap. Backward compatible: ok / reason / detail
+    // (e.message) unchanged; detail upgraded from plain string to object
+    // ONLY when stderr is present, so legacy callers reading detail as a
+    // string still get the truncated message via detail.message.
+    const message = String((e && e.message) || '').slice(0, 120);
+    const stderrRaw = (e && e.stderr) ? String(e.stderr) : '';
+    const diagnostic = stderrRaw.length > 200 ? stderrRaw.slice(-200) : stderrRaw;
+    const detail = stderrRaw.length > 0
+      ? { message: message, diagnostic: diagnostic }
+      : message;
     return {
       ok: false,
       reason: 'rs_engine_invocation_failed',
-      detail: String((e && e.message) || '').slice(0, 120),
+      detail: detail,
       pairs: [],
     };
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mindrian_os/install",
-  "version": "1.13.0-beta.28",
+  "version": "1.13.0-beta.30",
   "description": "Install MindrianOS into Claude Code with one command -- `npx @mindrian_os/install`. Ships the MindrianOS plugin (Larry + PWS methodology + Data Room) plus a setup/diagnostics CLI (install/doctor/update).",
   "scripts": {
     "mcp": "node bin/mindrian-mcp-server.cjs",