fix(research): preserve native Codex auth in isolated analysis

docs/reference/trajectory-assessment.md

@@ -21,6 +21,13 @@ the frontend does not present an unspecified or other-language result as a new
 analysis in the selected language. Output-language metadata records the request,
 not independent verification of a model's translation or judgments.
 
+The Codex observer reuses native Chat's shared authenticated HTTP transport.
+Credentials and refresh remain with Codex's existing signed-in account; the
+observer neither copies tokens nor imports user tool/rule configuration. Its
+explicit read-only sandbox, disabled shell/exec/web tools and fixed supplied
+evidence remain in force. A failed provider attempt stays in job history and
+requires an explicit retry after the host problem is repaired.
+
 ## What the measurements should answer
 
 Use one **research episode** as the process unit: a bounded question or competing

loopx/capabilities/benchmark_toolkit/trajectory_assessment.py

@@ -14,6 +14,7 @@
 from typing import Any
 
 from ...control_plane.effect_runtime import effect_runtime_result
+from ..native_chat.codex_context import shared_chatgpt_transport
 
 
 def _digest(value: Any) -> str:
@@ -144,12 +145,21 @@ def execute(
         _write(directory / "output-schema.json", prepared["output_schema"])
         (directory / "prompt.txt").write_text(prepared["prompt"], encoding="utf-8")
         config = prepared["config"]
+        # Keep native account routing while ignoring unrelated user tools and
+        # rules. The shared host transport delegates credentials/refresh to
+        # Codex; it supplies no endpoint, bearer token or account configuration.
+        transport = shared_chatgpt_transport({}, {})
+        provider = transport["model_provider"]
+        transport_args = ["-c", f"model_provider={json.dumps(provider)}"]
+        for key, value in transport["model_providers"][provider].items():
+            transport_args.extend(["-c", f"model_providers.{provider}.{key}={json.dumps(value)}"])
         command = [
             codex,
             "exec",
             "--ephemeral",
             "--ignore-user-config",
             "--ignore-rules",
+            *transport_args,
             "--skip-git-repo-check",
             "--sandbox",
             "read-only",

tests/capabilities/test_trajectory_assessment.py

@@ -91,6 +91,15 @@ def provider(
 import json,sys,time
 from pathlib import Path
 assert "\\u539f\\u59cb\\u76ee\\u6807" in sys.stdin.buffer.read().decode("utf-8")
+options = dict(arg.split('=', 1) for index, arg in enumerate(sys.argv) if index > 0 and sys.argv[index-1] == '-c')
+selected = json.loads(options['model_provider'])
+assert json.loads(options['model_providers.' + selected + '.requires_openai_auth']) is True
+assert json.loads(options['model_providers.' + selected + '.supports_websockets']) is False
+assert not any(key.endswith(('.base_url', '.env_key', '.experimental_bearer_token')) for key in options)
+assert '--ignore-user-config' in sys.argv and '--ignore-rules' in sys.argv
+assert sys.argv[sys.argv.index('--sandbox')+1] == 'read-only'
+assert json.loads(options['web_search']) == 'disabled'
+assert 'shell_tool' in sys.argv and 'unified_exec' in sys.argv
 time.sleep({delay})
 if {code}: sys.exit({code})
 Path(sys.argv[sys.argv.index('-o')+1]).write_bytes({payload!r})