omnius 1.0.847 → 1.0.849

package/docs/work-orders/evidence/native-conversation-prompt-audit-20261012.json

@@ -0,0 +1,226 @@
+{
+  "reviewedAtUtc": "2026-10-13T03:37:18.906666+00:00",
+  "runId": "telegram-c34a3d7d53728329-1791861851684-1",
+  "contextDumpSha256": "1d71cacae8554c37935c01127447b7e5e3983c2453085943eb4cd27ea7e9ef21",
+  "scope": "Sanitized prompt/tool-projection audit only; no full conversation/context, credentials, generated reasoning or raw payload bodies.",
+  "apiView": {
+    "kind": "model-facing-api-view",
+    "exactRequestStored": false,
+    "requestStorage": "redacted"
+  },
+  "toolNames": [
+    "file_read",
+    "web_fetch",
+    "web_search",
+    "telegram_media_acquire",
+    "file_write",
+    "file_edit",
+    "telegram_send_file",
+    "telegram_progress",
+    "task_complete",
+    "tool_search"
+  ],
+  "visibleCriteria": [
+    "effect:visible_response"
+  ],
+  "requiredMutationOrDeliveryCriteria": false,
+  "frontierAdviceAtTurn0": "recommended_action=No execution evidence is recorded yet. Choose the first feasible action that advances the user goal; use capability discovery when the required tool is deferred.",
+  "toolGroundTruthAtTurn0": "action_ground_truth=tool-derived only; controller artifacts and bookkeeping do not count as progress",
+  "projection": [
+    {
+      "tool": "file_write",
+      "descriptionChars": 121,
+      "requiredParameters": [
+        "path"
+      ],
+      "parameterNames": [
+        "path",
+        "content",
+        "content_base64",
+        "content_encoding",
+        "overwrite",
+        "expected_hash",
+        "dry_run"
+      ]
+    },
+    {
+      "tool": "telegram_send_file",
+      "descriptionChars": 121,
+      "requiredParameters": [
+        "path"
+      ],
+      "parameterNames": [
+        "path",
+        "target",
+        "chat_id",
+        "target_chat_id",
+        "user_id",
+        "username",
+        "kind",
+        "caption",
+        "reply_to_message_id"
+      ]
+    },
+    {
+      "tool": "task_complete",
+      "descriptionChars": 115,
+      "requiredParameters": [
+        "summary"
+      ],
+      "parameterNames": [
+        "summary",
+        "user_reply",
+        "deliver_artifacts",
+        "resume_delivery",
+        "status",
+        "question",
+        "missing_criterion_ids",
+        "closeout_reviewed",
+        "remaining_gaps",
+        "override_closeout_review",
+        "override_reason"
+      ]
+    }
+  ],
+  "toolSequence": [
+    {
+      "turn": 1,
+      "tool": "telegram_send_file",
+      "argumentNames": [
+        "content",
+        "reply_to_message_id",
+        "path",
+        "kind"
+      ],
+      "result": "scoped artifact missing",
+      "artifactBasename": "reply.txt"
+    },
+    {
+      "turn": 2,
+      "tool": "file_write",
+      "result": "created8bytes",
+      "artifactBasename": "reply.txt"
+    },
+    {
+      "turn": 3,
+      "tool": "task_complete",
+      "deliverArtifacts": false,
+      "result": "blocked:no real verification evidence"
+    },
+    {
+      "turn": 4,
+      "tool": "file_read"
+    },
+    {
+      "turn": 5,
+      "tool": "telegram_send_file",
+      "result": "provider upload"
+    },
+    {
+      "turn": 6,
+      "tool": "task_complete",
+      "result": "completed"
+    }
+  ],
+  "sourceFindings": [
+    {
+      "file": "packages/orchestrator/src/run-frontier.ts",
+      "finding": "No-observation branch prescribed first feasible action despite a response-only goal; repaired to allow direct conversation and limit tools/artifacts to goal material effects."
+    },
+    {
+      "file": "packages/orchestrator/src/agenticRunner.ts",
+      "line": 54966,
+      "finding": "General action_ground_truth tool-derived-only wording remains outside this narrow patch."
+    },
+    {
+      "file": "packages/orchestrator/src/completionContract.ts",
+      "line": 373,
+      "finding": "Generic execute/scenario/evidence/delivery flow remains outside this narrow patch."
+    },
+    {
+      "file": "packages/orchestrator/src/agenticRunner.ts",
+      "line": 56994,
+      "finding": "Tool descriptions take first120characters; long public workspace prefix obscured file_write purpose."
+    },
+    {
+      "file": "packages/cli/src/tui/telegram-bridge.ts",
+      "line": 3476,
+      "finding": "Native response contract explicitly permits inline-only answer without optional draft delivery."
+    },
+    {
+      "file": "packages/orchestrator/src/opencode-free-models.ts",
+      "finding": "Compatibility prefix/core read-shell metadata contains no instruction to create or upload a reply file."
+    }
+  ],
+  "laterRunFinalizationAudit": {
+    "runId": "telegram-c34a3d7d53728329-1791862080413-1",
+    "markerSha256": "1504c538cf090182adb37d293bcc417c098d8bce01f8fb861d58814dbfe92dd9",
+    "handlerResult": {
+      "timestamp": "2026-10-13T03:28:57.895Z",
+      "toolCallId": "call-3bda9728-0428-47ed-bceb-c63e466ceff7",
+      "argsHash": "844ca94b1d",
+      "success": true,
+      "semantic": "The real task_complete handler accepted the candidate; this alone is not a Telegram delivery acknowledgement.",
+      "userReplyChars": 2022,
+      "userReplySha256": "58638d481d65039bb41276c8d83878823d2132cff752d16102aa69ce31c81158",
+      "deliverArtifacts": false
+    },
+    "terminalProjection": {
+      "timestamp": "2026-10-13T03:28:57.988Z",
+      "toolCallId": null,
+      "argsHash": "077941984a",
+      "argsKeys": [
+        "terminalReceiptId",
+        "status",
+        "runtime"
+      ],
+      "status": "incomplete_verification",
+      "success": false,
+      "semantic": "Host-generated terminal receipt is recorded as task_complete/tool_result; it is not another invocation of the original tool call.",
+      "source": "packages/orchestrator/src/agenticRunner.ts:_commitCompletionFinalization debug projection"
+    },
+    "requiredEffects": [
+      "visible_response"
+    ],
+    "outcome": "conversation",
+    "observedEffects": [
+      "observation",
+      "mutation"
+    ],
+    "missingEffects": [
+      "visible_response"
+    ],
+    "unresolvedIds": [
+      "effect:visible_response"
+    ],
+    "unverifiedClaimIds": [],
+    "objectiveDecision": "deliver_due",
+    "deliveryReceiptsAtFinalization": 0,
+    "separateVerificationAudit": {
+      "status": "unverified",
+      "gaps": [
+        {
+          "id": "unresolved_0001",
+          "source": "verification_missing",
+          "detail": "File changes occurred without any successful verification evidence."
+        }
+      ]
+    },
+    "causalityLimits": [
+      "The visible-response obligation was still pending before host delivery; do not forge an ACK or mark it fulfilled from handler success.",
+      "The separate file-change verification audit is not sufficient by itself to prove why the runner selected incomplete_verification.",
+      "Actual later Telegram transport delivery must be checked independently of this pre-delivery terminal marker."
+    ],
+    "necessaryNextFixes": [
+      "Preserve a candidate/owner-bound ready-for-delivery handoff separately from actual fulfilled delivery, so an unsent host-owned visible response is not mislabeled failed execution verification.",
+      "Distinguish synthetic terminal-receipt projection from a real tool-call result in diagnostics; preserve both facts without appearing to execute one call twice.",
+      "Preserve concise tool purpose, chat scope and inline-reply guidance ahead of long workspace paths in model-visible metadata."
+    ]
+  },
+  "metadataRepair": {
+    "source": "packages/cli/src/tui/telegram-creative-tools.ts",
+    "scope": "Concise modelFacingSummary for scoped file_write/file_edit; exported scope-specific telegramSendFileModelFacingSummary for parent-owned Bridge integration.",
+    "authority": "No execution, path containment, artifact capability or Telegram target checks changed.",
+    "test": "Actual AgenticRunner tool projection with long workspace path plus denied outside-workspace write."
+  }
+}