omnius 1.0.699 → 1.0.700

package/docs/DISCOVERY.json

@@ -35173,7 +35173,7 @@
       "id": "guide.work-orders-runtime-health-remediation-wo-49-authored-final-delivery-uppercase",
       "kind": "guide",
       "title": "WO-49: Preserve the authored answer after tool work",
-      "summary": "Status: repository repair verified; scoped Git delivery in progress.",
+      "summary": "Status: repository repair complete; delivered as bfaabd7f to origin/main. Publication and live acceptance remain with the user.",
       "keywords": [
         "work",
         "orders",
@@ -35211,7 +35211,7 @@
       "id": "guide.work-orders-runtime-health-remediation-wo-50-embedding-unavailability-uppercase",
       "kind": "guide",
       "title": "WO-50: Recover honestly from missing embeddings",
-      "summary": "Status: source repair in progress; requested Nomic re-pull completed.",
+      "summary": "Status: repository repair complete; delivered as a3375c63 to origin/main. Requested Nomic re-pull completed and passive server recovery observed. Publication and live acceptance of source changes remain with the user.",
       "keywords": [
         "work",
         "orders",

package/docs/DISCOVERY.md

@@ -594,8 +594,8 @@ Daemon equivalents are `GET /v1/discovery/bootstrap`, `GET /v1/discovery?q=<inte
 | `guide.work-orders-runtime-health-remediation-wo-46-shell-result-authority-uppercase` | WO-46: Preserve process authority for shell observations | Status: repository repair complete and delivered to origin/main. Publication and live validation remain with the user. |
 | `guide.work-orders-runtime-health-remediation-wo-47-diagnostic-execution-contract-uppercase` | WO-47: Validate and execute diagnostics without blocking the agent | Status: repository repair complete and delivered to origin/main. Publication and live validation remain with the user. |
 | `guide.work-orders-runtime-health-remediation-wo-48-conversation-result-evidence-uppercase` | WO-48: Retain command evidence in conversation results | Status: repository repair complete and delivered to origin/main. Publication and live validation remain with the user. |
-| `guide.work-orders-runtime-health-remediation-wo-49-authored-final-delivery-uppercase` | WO-49: Preserve the authored answer after tool work | Status: repository repair verified; scoped Git delivery in progress. |
-| `guide.work-orders-runtime-health-remediation-wo-50-embedding-unavailability-uppercase` | WO-50: Recover honestly from missing embeddings | Status: source repair in progress; requested Nomic re-pull completed. |
+| `guide.work-orders-runtime-health-remediation-wo-49-authored-final-delivery-uppercase` | WO-49: Preserve the authored answer after tool work | Status: repository repair complete; delivered as bfaabd7f to origin/main. Publication and live acceptance remain with the user. |
+| `guide.work-orders-runtime-health-remediation-wo-50-embedding-unavailability-uppercase` | WO-50: Recover honestly from missing embeddings | Status: repository repair complete; delivered as a3375c63 to origin/main. Requested Nomic re-pull completed and passive server recovery observed. Publication and live acceptance of source changes remain with the user. |
 | `guide.work-orders-telegram-dmn-wo-22-dmn-outreach-and-learning-uppercase` | WO-22: DMN outreach, DM sharing, and outcome learning | On 2026-09-03 at 17:07 PDT the bot posted in the OMNIUS group without being addressed. The operator asked whether this was self-induced reflection. |
 | `guide.work-orders-telegram-dropbear-context-rca-workorder` | Telegram Dropbear Context Engineering RCA Work Order | Observed run: /home/roko/Documents/Projects/Adjacent/telegramtest/.omnius, run id 1782873796963-i5r7mv. |
 | `guide.work-orders-wo-am-gaps-uppercase` | Associative Memory Gap Work Orders | Generated: 2026-04-13 Source: Deep audit of multimodal associative memory systems Status: READY FOR IMPLEMENTATION |

package/docs/work-orders/runtime-health-remediation/TRACKER.md

@@ -6,8 +6,8 @@
 
 ## September 5 runtime 1.0.698 delivery and Ollama follow-up
 
-- [ ] [WO-49: authored final delivery](WO-49-authored-final-delivery.md): generated answer lost after earlier prose and tool work; source repair and real-runner Telegram regression in progress.
-- [ ] [WO-50: embedding unavailability](WO-50-embedding-unavailability.md): bounded missing-model recovery, truthful vector/job outcomes and exact model identity in progress. User-requested Nomic re-pull succeeded; installation metadata verified without inference.
+- [x] [WO-49: authored final delivery](WO-49-authored-final-delivery.md): later authored replies survive earlier prose, tool work and summary-only completion. Commit `bfaabd7f` delivered to `origin/main`; 280 orchestrator and 58 CLI checks passed.
+- [x] [WO-50: embedding unavailability](WO-50-embedding-unavailability.md): commit `a3375c63` delivered to `origin/main`; bounded missing-model recovery, truthful vector/job outcomes and exact request-model identity verified. User-requested Nomic re-pull succeeded; passive journal follow-up recorded 51 embedding HTTP 200 responses and no embedding 404s in the captured post-pull window. Clean all-workspace build passed, memory 665/665 and indexer 186/186 passed, twelve additional dependent CLI cases and all seventeen daemon transport cases passed. The three real Telegram answer-delivery cases also passed after the clean rebuild. Publication and live acceptance of source changes remain with the user; existing-data/model-migration limits are explicit in WO-50.
 
 The separate 14:43 PDT router failure was a managed Ollama lane exiting mid-stream, followed by connection refused and recovery. The external broker does not retain its exit status/stderr, so the underlying exit cause remains unresolved; WO-50 records the evidence and external observability follow-up.
 

package/docs/work-orders/runtime-health-remediation/WO-49-authored-final-delivery.md

@@ -1,6 +1,6 @@
 # WO-49: Preserve the authored answer after tool work
 
-Status: repository repair verified; scoped Git delivery in progress.
+Status: repository repair complete; delivered as `bfaabd7f` to `origin/main`. Publication and live acceptance remain with the user.
 
 ## Field evidence and cause
 
@@ -23,7 +23,7 @@ Selected evidence is copied with SHA-256 manifests under `/tmp/omnius-run-178864
 - [x] Stream and batch paths tag actual model text consistently; a response emits exactly once.
 - [x] Both loops retain interruption ownership; pause holds output, resume emits once, Stop rejects late output.
 - [x] Intervening work and task replacement invalidate stale answers; unvalidated URLs remain subject to the existing evidence gate.
-- [ ] Record verification and scoped commit/push.
+- [x] Record verification and scoped commit/push.
 
 ## Limits
 
@@ -37,3 +37,5 @@ The copied `•` bullets were not reproduced in repository formatting or JSO
 - `OMNIUS_SQLITE_TESTS=1 pnpm --dir packages/orchestrator exec vitest run tests/agenticRunner-interruption.test.ts tests/structured-tool-authority.test.ts tests/agenticRunner.test.ts tests/conversation-report-runner.test.ts --maxWorkers=4 --minWorkers=1`: 276 passed across four suites. Combined with the four new producer cases: 280 distinct orchestrator checks. Log: `/tmp/omnius-wo49-orchestrator-focused.log`.
 - `pnpm --dir packages/cli exec vitest run tests/telegram-terminal-delivery.test.ts tests/telegram-streamed-terminal-answer.test.ts tests/telegram-conversation-evidence.test.ts tests/telegram-stream-timeout.test.ts --maxWorkers=4 --minWorkers=1`: 58 passed across four suites. Log: `/tmp/omnius-wo49-cli-focused.log`.
 - Orchestrator build passed (`/tmp/omnius-wo49-orchestrator-build.log`). All inference and Telegram delivery in these tests were mocked. Publication and live acceptance remain with the user.
+
+Final integration also passed a clean all-workspace rebuild and all 17 required daemon transport/client/script cases; logs are recorded in WO-50.

package/docs/work-orders/runtime-health-remediation/WO-50-embedding-unavailability.md

@@ -1,6 +1,6 @@
 # WO-50: Recover honestly from missing embeddings
 
-Status: source repair in progress; requested Nomic re-pull completed.
+Status: repository repair complete; delivered as `a3375c63` to `origin/main`. Requested Nomic re-pull completed and passive server recovery observed. Publication and live acceptance of source changes remain with the user.
 
 ## Field evidence and operational remedy
 
@@ -19,15 +19,33 @@ Passive journal follow-up through 15:25:50 PDT then recorded 51 `/api/embed` HTT
 
 ## Acceptance
 
-- [ ] Missing-model bursts make bounded requests, with no implicit downloads or recursive retries.
-- [ ] Exact endpoint/model identity, TTL recovery and independent cancellation are tested.
-- [ ] Configuration honors the selected model; availability checks cannot match a different tag by prefix.
-- [ ] Failed generation cannot fabricate vectors or complete embedding jobs; text remains recoverable.
-- [ ] Compatibility handling cannot fan one model failure into one request per input.
-- [ ] Record verification and scoped commit/push.
+- [x] Missing-model bursts make bounded requests, with no implicit downloads or recursive retries.
+- [x] Exact endpoint/model identity, TTL recovery and independent cancellation are tested.
+- [x] Configuration honors the selected model; availability checks cannot match a different tag by prefix.
+- [x] Failed generation cannot fabricate vectors or complete embedding jobs; text remains recoverable.
+- [x] Compatibility handling cannot fan one model failure into one request per input.
+- [x] Record verification and scoped commit/push.
 
 ## Separate chat-server incident
 
 This embedding defect did not cause the observed router HTTP 503. At 14:43:43.457 PDT the broker logged an incomplete `/api/chat` read; at 14:43:43.469 its fresh retry hit connection refused. At 14:43:44.583 it reported `lane-121` stopped unexpectedly. The lane had been ready since 14:25:42 for `robit/qwen3.8-27b-obliterated-e03:27b` on GPU UUID `GPU-170a99ee-850f-2182-1050-4e8d3c87b6b0`. The broker served chat again at 14:44:09.532, and the main run subsequently produced its answer.
 
 No OOM kill, GPU Xid, segfault or coredump was found in the narrow incident journal. The external broker discards child stdout/stderr and omits PID/exit status in its prune warning (`/usr/local/libexec/ollama-unify-gpu-negotiator`, child spawn and `_prune_dead_lanes_locked`). Its exact exit mechanism remains unknown. An external broker follow-up needs retained child stderr and return-code logging to distinguish crash, signal and deliberate exit; this Omnius repair does not claim to fix that unknown cause. Hashed journal captures are under `/tmp/omnius-ollama-failure-aki8oz5k/`; the earlier incident window is under `/tmp/omnius-ollama-early-failure-_nsbctcb/`.
+
+## Review corrections and existing data limits
+
+Review added regression coverage for preserving a custom embedding set through immediate failure and queued retry, for blank environment overrides, for numerically unusable indexer vectors, and for diagnostics that omit provider payloads and URL credentials.
+
+The 30-second availability cooldown is process-local and bounded to 128 endpoint/model identities. It expires after an operator installs the model. These changes do not requeue historical jobs already incorrectly marked complete, re-embed existing vectors, or migrate a stored vector space when the embedding model changes. Restoring the same Nomic model does not require choosing a substitute embedding space.
+
+The availability key is not a guarantee of stored vector-space isolation. A separate existing model-change limitation spans `UnifiedMemoryStore.searchVector` (dimension/set filters without query model), `EpisodeStore.search/setEmbedding` (no persisted model identity), and `ProceduralMemoryStore.searchSemantic` (stored model not used for comparison). Any future model-migration repair must preserve model identity from generation through storage and query together: unified callers in execution `memory-search.ts` and CLI `routes-v1.ts`; episode callers in `agenticRunner.ts` and `telegram-reflection-corpus.ts`; procedural callers in `proceduralMemoryStore.ts`. It needs exact-tag/unknown-model tests in the corresponding store and caller suites and lexical fallback for incompatible vectors. No partial schema migration or automatic re-embedding was attempted in this missing-model repair.
+
+## Final verification
+
+- `OMNIUS_SQLITE_TESTS=1 pnpm --dir packages/memory exec vitest run`: 665 passed across all 55 suites, zero skips. Includes sequential/concurrent missing-model bursts, exact identities and expiry, independent cancellation, malformed vectors, custom-set retry, raw-text retrieval, and blank environment overrides. Log: `/tmp/omnius-wo50-memory-full-final.log`.
+- `pnpm --dir packages/indexer exec vitest run`: 186 passed across all 13 suites. The 39 provider cases cover missing-model single-request failure, genuine legacy-route compatibility, no fabricated vectors, exact dimensions/cardinality, numeric overflow/underflow, configuration, and redacted timeout/cancellation errors. Log: `/tmp/omnius-indexer-embedding-all.log`.
+- Both package typechecks/builds passed. A subsequent clean rebuild of every workspace package passed after deleting stale build metadata, without running the publish bundler or changing the user's publish staging. Log: `/tmp/omnius-wo49-50-clean-build.log`.
+- After that rebuild, actual Telegram authored-answer integration and the reflection corpus/maintenance suites passed all 15 checks (three repeat WO-49 cases and twelve dependent integration cases). Log: `/tmp/omnius-wo49-50-cli-integration.log`.
+- All 17 required daemon transport/client/script cases passed, including split streams, delayed errors, subprocess failures and managed Stop. Log: `/tmp/omnius-wo49-frontend-transport.log`.
+
+All inference and Telegram HTTP in repository checks was mocked. The sole model operation on the live server was the user-requested Nomic pull; subsequent live embedding success was observed passively in the journal. No live test inference, service restart, installed-package replacement, Telegram sends or publication occurred.

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.699",
+  "version": "1.0.700",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.699",
+      "version": "1.0.700",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.699",
+  "version": "1.0.700",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/library.js",