Fix strict cost accounting for verified Copilot context parser refusals

argus_skill/adapters/agent_cli_backend/_exec_finalize.py

@@ -39,6 +39,7 @@ def finalize_result(
     token_usage: TokenUsage | None = None,
     premium_requests: float | None = None,
     error: str = "",
+    startup_receipt: dict | None = None,
 ) -> RunnerResult:
     backend = ctx.backend
     persisted_error = redact_secrets_text(
@@ -161,6 +162,7 @@ def finalize_result(
                 thread_id=result.thread_id,
                 model_usage=result.model_usage,
                 error=persisted_error,
+                startup_receipt=startup_receipt,
             )
             appended = UsageLedger(
                 ctx.usage_project_root,

argus_skill/adapters/agent_cli_backend/_exec_spawn.py

@@ -414,4 +414,5 @@ def spawn_and_finish(ctx: "_ExecContext", cli_options: Any) -> RunnerResult:
             else "completed"
         ),
         error=safe_failure_text,
+        startup_receipt=complete_row,
     )

argus_skill/core/runner_errors.py

@@ -126,3 +126,51 @@ def result_has_pre_provider_refusal(result: Any) -> bool:
     "result_has_pre_provider_refusal",
     "result_has_unrecoverable_resume_state",
 ]
+
+
+def is_copilot_context_parser_error(value: object) -> bool:
+    """Exact runner-wrapped parser diagnostic; text alone is not authority."""
+    prefix = (
+        "Process exited with code 1 before turn completion.\n"
+        "error: unknown option '--context'\n"
+    )
+    suffix = "Try 'copilot --help' for more information."
+    return str(value or "").strip() in (
+        prefix + "(Did you mean --connect?)\n\n" + suffix,
+        prefix + "\n" + suffix,
+    )
+
+
+def is_copilot_context_parser_refusal(
+    error: object, *, provider: str, call_id: str, run_label: str,
+    status: str, thread_id: object, source: str,
+    receipt: dict[str, Any] | None,
+) -> bool:
+    """Use only host-generated agent.io.complete, never model/tool JSON.
+
+    A parser diagnostic is positive startup evidence only when the matching
+    process receipt confirms an unsuccessful, silent pre-turn CLI invocation.
+    Usage accounting must independently reject every observed usage field.
+    """
+    if not is_copilot_context_parser_error(error) or not receipt:
+        return False
+    command = receipt.get("command")
+    return bool(
+        provider == "copilot" and status == "error" and source == "run_exec"
+        and not thread_id and call_id
+        and receipt.get("type") == "agent.io.complete"
+        and receipt.get("backend") == provider
+        and receipt.get("call_id") == call_id
+        and receipt.get("run_label") == run_label
+        and receipt.get("exit_code") == 1
+        and receipt.get("turn_failed") is True
+        and receipt.get("turn_completed") is False
+        and receipt.get("thread_id") is None
+        and receipt.get("fatal_error") == "Process exited with code 1 before turn completion."
+        and receipt.get("tool_activity_observed") is False
+        and all(receipt.get(key) == 0 for key in (
+            "agent_message_count", "stdout_line_count", "json_event_count"))
+        and isinstance(command, list)
+        and any(command[i:i+2] == ["--context", "default"]
+                for i in range(1, len(command)-1))
+    )

argus_skill/core/usage.py

@@ -26,7 +26,11 @@
 )
 from .event_catalog import CALL_SCOPED_EVENT_TYPES, EventType, canonical_event_type
 from .pricing import PricingQuote, PricingStatus, quote_copilot_usage, quote_token_usage
-from .runner_errors import is_pre_provider_refusal_error
+from .runner_errors import (
+    is_copilot_context_parser_error,
+    is_copilot_context_parser_refusal,
+    is_pre_provider_refusal_error,
+)
 from .token_usage import TokenUsage, extract_token_usage
 
 try:  # pragma: no cover - Windows usage mutations use portalocker below
@@ -90,13 +94,26 @@ def to_jsonable(self) -> dict[str, Any]:
         return row
 
     @classmethod
-    def from_jsonable(cls, row: dict[str, Any]) -> "UsageRecord":
+    def from_jsonable(
+        cls, row: dict[str, Any], *, startup_receipt: dict[str, Any] | None = None,
+    ) -> "UsageRecord":
         cost = _optional_float(row.get("cost_usd"))
         pricing_status = _pricing_status(row.get("pricing_status"))
         pricing_tier = str(row.get("pricing_tier") or "unknown")
         error = str(row.get("error") or "")
         if (
-            is_pre_provider_refusal_error(error)
+            (is_pre_provider_refusal_error(error) or (
+                is_copilot_context_parser_refusal(
+                    error, provider=str(row.get("provider") or ""),
+                    call_id=str(row.get("call_id") or ""),
+                    run_label=str(row.get("run_label") or ""),
+                    status=str(row.get("status") or ""),
+                    thread_id=row.get("thread_id"),
+                    source=str(row.get("source") or ""), receipt=startup_receipt,
+                )
+                and row.get("premium_requests") is None
+                and row.get("premium_request_cost_usd") is None
+            ))
             and cost is None
             and row.get("total_nano_aiu") is None
             and not row.get("model_usage")
@@ -272,13 +289,20 @@ def build_usage_record(
     model_usage: Iterable[dict[str, Any]] | None = None,
     error: str = "",
     source: UsageSource = "run_exec",
+    startup_receipt: dict[str, Any] | None = None,
 ) -> UsageRecord:
     usage = token_usage or TokenUsage()
     normalized_model_usage = _normalize_model_usage(model_usage)
     normalized_provider = str(provider or "").strip().lower()
     premium_quote = quote_copilot_usage(premium_requests)
     missing_resume_target = (
-        is_pre_provider_refusal_error(error)
+        (is_pre_provider_refusal_error(error) or (
+            is_copilot_context_parser_refusal(
+                error, provider=normalized_provider, call_id=call_id,
+                run_label=run_label, status=status, thread_id=thread_id,
+                source=source, receipt=startup_receipt,
+            ) and premium_requests is None
+        ))
         and total_nano_aiu is None
         and provider_cost_usd is None
         and not normalized_model_usage
@@ -495,6 +519,7 @@ def records(
             handle = self.path.open("r", encoding="utf-8")
         except OSError:
             return out
+        startup_receipts = None
         with handle:
             for raw in handle:
                 try:
@@ -503,7 +528,12 @@ def records(
                     continue
                 if not isinstance(row, dict):
                     continue
-                record = UsageRecord.from_jsonable(row)
+                receipt = None
+                if is_copilot_context_parser_error(row.get("error")):
+                    if startup_receipts is None:
+                        startup_receipts = _startup_completion_receipts(self.project_root)
+                    receipt = startup_receipts.get(str(row.get("call_id") or ""))
+                record = UsageRecord.from_jsonable(row, startup_receipt=receipt)
                 if not record.call_id or record.call_id in seen:
                     continue
                 seen.add(record.call_id)
@@ -1689,3 +1719,31 @@ def _call_status(value: Any) -> CallStatus:
     "summarize_usage",
     "usage_recorded_event",
 ]
+
+
+def _startup_completion_receipts(project_root: Path) -> dict[str, dict[str, Any]]:
+    """Read host lifecycle summaries, not raw provider/tool stream frames.
+
+    Ambiguous duplicate completion receipts fail closed. Only consulted when
+    an exact historical parser diagnostic needs the missing runner context.
+    """
+    receipts: dict[str, dict[str, Any]] = {}
+    seen: set[str] = set()
+    try:
+        with (project_root / "events.jsonl").open(encoding="utf-8") as handle:
+            for raw in handle:
+                try:
+                    event = json.loads(raw)
+                except (ValueError, TypeError):
+                    continue
+                if not isinstance(event, dict) or event.get("type") != "agent.io.complete":
+                    continue
+                call_id = str(event.get("call_id") or "")
+                if call_id in seen:
+                    receipts.pop(call_id, None)
+                else:
+                    receipts[call_id] = event
+                seen.add(call_id)
+    except OSError:
+        return {}
+    return receipts

argus_skill/release_manifest.json

@@ -1,6 +1,6 @@
 {
   "package_version": "0.1.1",
-  "release_id": "0.1.1+b9d192ba6b186634",
+  "release_id": "0.1.1+f2f2ab05a41299fc",
   "schema_version": 1,
-  "source_digest": "b9d192ba6b1866346496b561464a2052d261b03b2eab1424d2299f676eb9ba13"
+  "source_digest": "f2f2ab05a41299fc21fbb430c0bc5de704d2c01259d39ae5bfac6a98489bf776"
 }

frontend/core/src/release.generated.ts

@@ -1,3 +1,3 @@
 // Generated by argus_skill.release_tools.generate_manifest. Do not edit.
-export const RELEASE_ID = "0.1.1+b9d192ba6b186634";
-export const RELEASE_SOURCE_DIGEST = "b9d192ba6b1866346496b561464a2052d261b03b2eab1424d2299f676eb9ba13";
+export const RELEASE_ID = "0.1.1+f2f2ab05a41299fc";
+export const RELEASE_SOURCE_DIGEST = "f2f2ab05a41299fc21fbb430c0bc5de704d2c01259d39ae5bfac6a98489bf776";