omnius 1.0.765 → 1.0.766

package/dist/index.js

@@ -1,4 +1,4 @@
 [diffend] Oversized file quarantined before diffing.
 name: package/dist/index.js
-size: 38754141 bytes
-sha256: 8f6cc23abacc45c47703361d743118fe124f196c880cd3f05c49f8b8320609a9
+size: 38763840 bytes
+sha256: 3199bdafa0335c954b90a33a7627b68596201ba4fb4f11880d72d9ba5debd567

package/docs/DISCOVERY.json

@@ -35751,6 +35751,44 @@
         }
       ]
     },
+    {
+      "id": "guide.work-orders-runtime-health-remediation-wo-55-telegram-pressure-recovery-uppercase",
+      "kind": "guide",
+      "title": "WO-55: Telegram pressure recovery without silent terminal states",
+      "summary": "Status: complete Priority: P0 Incident: September 26, 2026, Case Theory messages 458-461 Owner: current implementation session",
+      "keywords": [
+        "work",
+        "orders",
+        "runtime",
+        "health",
+        "remediation",
+        "WO",
+        "55",
+        "telegram",
+        "pressure",
+        "recovery",
+        "md"
+      ],
+      "maturity": "internal",
+      "audiences": [
+        "maintainer",
+        "large-context-agent"
+      ],
+      "layer": "documentation",
+      "interfaces": [
+        {
+          "type": "file",
+          "target": "docs/work-orders/runtime-health-remediation/WO-55-telegram-pressure-recovery.md"
+        }
+      ],
+      "references": [
+        {
+          "type": "documentation",
+          "target": "docs/work-orders/runtime-health-remediation/WO-55-telegram-pressure-recovery.md",
+          "relation": "canonical-artifact"
+        }
+      ]
+    },
     {
       "id": "guide.work-orders-telegram-dmn-wo-22-dmn-outreach-and-learning-uppercase",
       "kind": "guide",

package/docs/DISCOVERY.md

@@ -606,6 +606,7 @@ Daemon equivalents are `GET /v1/discovery/bootstrap`, `GET /v1/discovery?q=<inte
 | `guide.work-orders-runtime-health-remediation-wo-52-steering-projection-preservation-uppercase` | WO-52: Preserve required steering instructions in the actual model request | Status: source repair complete; verified and ready for publication. Live acceptance follows the user's publication. |
 | `guide.work-orders-runtime-health-remediation-wo-53-deferred-reflection-churn-uppercase` | WO-53: Stop duplicate deferred reflection writes | Status: scoped source repair verified; publication remains with the operator. Issue: https://github.com/robit-man/open-agents/issues/3 |
 | `guide.work-orders-runtime-health-remediation-wo-54-case-theory-router-deferral-recovery-uppercase` | WO-54: Case Theory router deferral recovery | Status: complete Priority: P0 Incident: September 26, 2026, Case Theory messages 455–456 Runtime observed: globally installed Omnius 1.0.764 Owner: current implementation session |
+| `guide.work-orders-runtime-health-remediation-wo-55-telegram-pressure-recovery-uppercase` | WO-55: Telegram pressure recovery without silent terminal states | Status: complete Priority: P0 Incident: September 26, 2026, Case Theory messages 458-461 Owner: current implementation session |
 | `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

@@ -4,6 +4,10 @@
 **Checked-item rule:** code, focused tests, and named evidence must all exist
 **Last reconciled:** 2026-09-05
 
+## September 26 Case Theory pressure recovery
+
+- [x] [WO-55: Telegram pressure recovery without silent terminal states](WO-55-telegram-pressure-recovery.md): generation-fenced recovery for silent inference, stale active-runner decisions, terminal review, continuation, and exact Telegram delivery receipts.
+
 ## September 26 Case Theory router deferral recovery
 
 - [x] [WO-54: Case Theory router deferral recovery](WO-54-case-theory-router-deferral-recovery.md): exact task-identity retirement, sequential mixed-domain repair, one scheduler-owned recovery with continuous typing and truthful exhaustion, plus canonical receipt-backed URLs with inference review and final fail-closed withholding. Implemented in `d6a6cb6e`; 251 focused Telegram tests and CLI build passed.

package/docs/work-orders/runtime-health-remediation/WO-55-telegram-pressure-recovery.md

@@ -0,0 +1,122 @@
+# WO-55: Telegram pressure recovery without silent terminal states
+
+**Status:** complete
+**Priority:** P0
+**Incident:** September 26, 2026, Case Theory messages 458-461
+**Owner:** current implementation session
+
+## Objective
+
+Keep one inference-authorized Telegram turn recoverable across broker pressure,
+silent model streams, invalid router contracts, terminal-review unavailability,
+and retryable Bot API failures. Recovery must retain the exact intake and exact
+generated payload, remain generation-fenced, and eventually return to inference
+or delivery without manufacturing canned replies or host-authored semantic
+decisions.
+
+## Confirmed root-cause chain
+
+1. Case Theory admitted a direct research follow-up and ran a long tool-backed
+   foreground task under substantial model/context pressure.
+2. The runner produced substantive final text and crossed its internal terminal
+   boundary before ordinary Telegram text transport ran.
+3. Runner `visible_response` evidence describes generated model-visible text;
+   the ordinary Telegram message receipt is recorded later in a separate turn
+   delivery ledger. The two lifecycles were presented as one completed turn
+   even when no Telegram message ID existed.
+4. Retryable/ambiguous delivery ledger states returned `null`, after which the
+   outer path could label the activity complete without consuming the reply
+   owner or scheduling exact-payload recovery.
+5. A fresh terminal metacognitive review returning no typed contract set the
+   intake to `deferred` but deliberately scheduled no continuation.
+6. Incomplete terminal receipts and pre-terminal runner errors used the same
+   deferred-without-work pattern.
+7. The metacognitive continuation counter became a terminal three-cycle cap
+   even when fresh inference still selected `continue`.
+8. Router recovery made one scheduler replay, then converted another invalid or
+   unavailable contract into a terminal failed intake.
+9. A follow-up routed while the old runner was settling could receive a
+   no-reply decision against the old active-run snapshot. If the runner exited
+   before that decision returned, the bridge consumed the follow-up instead of
+   re-running inference against the new terminal/open-task state.
+10. The stale-sub-agent watchdog explicitly exempted `waitingForInference`,
+    while Telegram action runners set `requestTimeoutMs: 0`. A silent admitted
+    stream could therefore pin its chat, typing lease, queue/admission slot, and
+    inference-pressure lease forever.
+11. Late settlement cleanup was not identity-conditional, so simply freeing a
+    wedged generation could let its eventual `finally` delete newer generation
+    state.
+
+## Required invariants
+
+- Reply authority, continuation, completion, and no-reply remain inference
+  decisions. Lifecycle recovery may retry an unresolved decision but cannot
+  invent one.
+- No generated acknowledgment or canned progress reply is introduced.
+- Every nonterminal intake has concrete scheduled work, an active generation,
+  or an explicit recoverable delivery-reconciliation state.
+- A router decision made for a runner that settled before the decision returned
+  is stale control evidence and must be inferred again.
+- Broker/capacity and typed-contract failures retry through the bounded session
+  scheduler with capped backoff; they never spin detached inference loops.
+- A fresh metacognitive `continue` decision is obeyed regardless of how many
+  earlier cycles ran. The counter controls backoff/telemetry, not semantics.
+- Silent inference has a bounded progress epoch. Expiry aborts and fences the
+  old generation, releases local admission, and requeues the exact intake.
+- Late events and cleanup from a retired generation cannot mutate or delete its
+  replacement.
+- Typing is present while a user-facing generation or recovery attempt is
+  actively admitted, and is stopped during true cancellation or terminal
+  delivery.
+- Known pre-acceptance/rate-limit Telegram failures retry the same immutable
+  delivery operation. A stored accepted message ID is locally reconciled to a
+  receipt without resending.
+- Transport ambiguity is never blindly resent. It becomes an explicit durable
+  reconciliation state with the intake and payload retained, rather than a
+  false completion or invisible wait.
+
+## Implementation checklist
+
+- [x] Make router recovery durable and backoff-scheduled instead of terminally
+      exhausted.
+- [x] Re-run stale active-runner no-reply decisions after runner settlement.
+- [x] Requeue terminal-review, incomplete-terminal, and safe pre-action fatal
+      paths instead of writing inert `deferred` records.
+- [x] Remove the deterministic metacognitive continuation stop; retain bounded
+      scheduler backoff and observability.
+- [x] Retire silent admitted inference after a bounded meaningful-progress
+      epoch and generation-fence every late callback/finalizer.
+- [x] Make old-generation cleanup conditional on ownership of the current map
+      entry.
+- [x] Reconcile delivery records that already contain an accepted Telegram
+      message ID without another network send.
+- [x] Retry known pre-acceptance/rate-limit failures using the same immutable
+      delivery operation and payload.
+- [x] Keep ambiguous delivery explicit and durable for operator reconciliation
+      without blind resend or false completion.
+- [x] Merge a recovering router snapshot behind newer same-session input so
+      neither intake overwrites the other and the newest retains decision
+      ownership.
+- [x] Add deterministic regressions for every recovery boundary above.
+- [x] Run focused Telegram lifecycle, intake, terminal-delivery, transport,
+      continuity, and typing suites.
+- [x] Run the CLI build and workspace build, review the scoped diff, commit, and
+      push without touching publication/version files.
+
+## Verification evidence
+
+- `pnpm --filter @omnius/orchestrator exec vitest run tests/inference-pressure.test.ts --reporter=dot`
+  passed 9/9 tests, including immediate lease release when an unresponsive
+  backend is aborted.
+- The focused Telegram command covering steering finalization, Bot API routing,
+  lifecycle leak prevention, delivery continuity, transport contracts,
+  terminal delivery, and working indicators passed 263/263 tests across seven
+  files.
+- `pnpm -r build` passed all 11 selected workspace projects, including the
+  orchestrator, CLI, API, and worker TypeScript builds.
+- Deterministic regressions cover stale no-reply inference, silent admitted
+  generation retirement, exact-intake recovery, continuation past cycle three,
+  concurrent newer-intake merging, 429 exact-payload retry, accepted-receipt
+  restart reconciliation, and process pressure-lease release.
+- No canned progress response, deterministic semantic reply gate, npm publish,
+  global install, daemon restart, or live Telegram mutation was added or run.

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.765",
+  "version": "1.0.766",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.765",
+      "version": "1.0.766",
       "bundleDependencies": [
         "image-to-ascii"
       ],
@@ -2664,9 +2664,9 @@
       }
     },
     "node_modules/ansi-regex": {
-      "version": "6.3.0",
-      "resolved": "https://registry.npmjs.org/ansi-regex/-/ansi-regex-6.3.0.tgz",
-      "integrity": "sha512-WpDfL7NO6j7tH88IDBNVdUJxDh9nmCteAVW9dsep846XdwF4naCBK+/tGLX3KJgcpgMRXCFlTM2hKGoK9FsdrQ==",
+      "version": "6.4.0",
+      "resolved": "https://registry.npmjs.org/ansi-regex/-/ansi-regex-6.4.0.tgz",
+      "integrity": "sha512-KzTVk2tCWAHtYrvvvaP8bJKJq2pVinhLcGEQdtLIYPbmNGNyYe8QwNaTUYQp2J7/vIsUKt5QCqAfUkYyG9DkOw==",
       "license": "MIT",
       "engines": {
         "node": ">=12"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.765",
+  "version": "1.0.766",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/library.js",