hermes(agent): reuse Codex CLI auth for auxiliaries

services/hermes/agent-deployment.yaml

@@ -24,7 +24,7 @@ spec:
         ai.bstein.dev/execution: Hermes Kanban with durable direct Codex and Claude Code CLI workers
         ai.bstein.dev/model-policy: Jetson-assisted AUTO routing, low through xhigh, cross-provider fallback
         ai.bstein.dev/placement: rpi5 preferred; Jetson deferred until state storage is available
-        ai.bstein.dev/config-rev: "20260811-codex-cross-provider-fallback"
+        ai.bstein.dev/config-rev: "20260811-codex-cli-auxiliary"
         vault.hashicorp.com/agent-inject: "true"
         vault.hashicorp.com/role: hermes-agent
         vault.hashicorp.com/agent-inject-secret-anthropic-token: kv/data/atlas/hermes/agent-tokens
@@ -279,6 +279,8 @@ spec:
             - /patched/chat_completion_helpers.py
             - /opt/hermes/agent/conversation_loop.py
             - /patched/conversation_loop.py
+            - /opt/hermes/agent/auxiliary_client.py
+            - /patched/auxiliary_client.py
           securityContext:
             allowPrivilegeEscalation: false
             runAsUser: 10000
@@ -424,6 +426,7 @@ spec:
             - {name: codex-runtime-patch, mountPath: /opt/hermes/agent/codex_runtime.py, subPath: codex_runtime.py}
             - {name: codex-runtime-patch, mountPath: /opt/hermes/agent/chat_completion_helpers.py, subPath: chat_completion_helpers.py}
             - {name: codex-runtime-patch, mountPath: /opt/hermes/agent/conversation_loop.py, subPath: conversation_loop.py}
+            - {name: codex-runtime-patch, mountPath: /opt/hermes/agent/auxiliary_client.py, subPath: auxiliary_client.py}
             - {name: tui-gateway-patch, mountPath: /opt/hermes/tui_gateway/server.py, subPath: server.py}
             - {name: kubeconfig, mountPath: /opt/data/home/.kube/config, subPath: config, readOnly: true}
             - {name: auto-router-plugin, mountPath: /opt/data/plugins/auto-router, readOnly: true}
@@ -574,6 +577,7 @@ spec:
             - {name: codex-runtime-patch, mountPath: /opt/hermes/agent/codex_runtime.py, subPath: codex_runtime.py}
             - {name: codex-runtime-patch, mountPath: /opt/hermes/agent/chat_completion_helpers.py, subPath: chat_completion_helpers.py}
             - {name: codex-runtime-patch, mountPath: /opt/hermes/agent/conversation_loop.py, subPath: conversation_loop.py}
+            - {name: codex-runtime-patch, mountPath: /opt/hermes/agent/auxiliary_client.py, subPath: auxiliary_client.py}
             - {name: tui-gateway-patch, mountPath: /opt/hermes/tui_gateway/server.py, subPath: server.py}
             - {name: kubeconfig, mountPath: /opt/data/home/.kube/config, subPath: config, readOnly: true}
             - {name: auto-router-plugin, mountPath: /opt/data/plugins/auto-router, readOnly: true}

services/hermes/scripts/patch_codex_runtime.py

@@ -223,6 +223,50 @@ FALLBACK_CONTEXT_AFTER = '''def build_cross_provider_codex_prompt(
 def run_codex_app_server_turn(
 '''
 
+AUXILIARY_TOKEN_BEFORE = '''    except Exception as exc:
+        logger.debug("Could not read Codex auth for auxiliary client: %s", exc)
+        return None
+'''
+AUXILIARY_TOKEN_AFTER = '''    except Exception as exc:
+        logger.debug("Could not read Hermes Codex auth for auxiliary client: %s", exc)
+
+        # Agent Hermes deliberately runs Codex through the authenticated CLI
+        # app-server and does not duplicate those credentials into Hermes'
+        # provider auth store. Auxiliary tasks still use Hermes' native Codex
+        # Responses adapter, so read the current CLI access token without
+        # copying or refreshing it here. The Codex CLI remains the sole owner
+        # of refresh-token rotation.
+        try:
+            codex_home = os.environ.get("CODEX_HOME", "").strip()
+            if not codex_home:
+                codex_home = str(Path.home() / ".codex")
+            auth_path = Path(codex_home).expanduser() / "auth.json"
+            payload = json.loads(auth_path.read_text(encoding="utf-8"))
+            tokens = payload.get("tokens") or {}
+            access_token = tokens.get("access_token")
+            if not isinstance(access_token, str) or not access_token.strip():
+                return None
+
+            # Match the native expiry check above. An expired CLI token is not
+            # refreshed from this side channel because that would race the
+            # app-server's canonical refresh-token owner.
+            try:
+                import base64
+                jwt_payload = access_token.split(".")[1]
+                jwt_payload += "=" * (-len(jwt_payload) % 4)
+                claims = json.loads(base64.urlsafe_b64decode(jwt_payload))
+                expires_at = claims.get("exp", 0)
+                if expires_at and time.time() > expires_at:
+                    logger.debug("Codex CLI access token is expired; skipping auxiliary route")
+                    return None
+            except Exception:
+                pass
+            return access_token.strip()
+        except Exception as cli_exc:
+            logger.debug("Could not read Codex CLI auth for auxiliary client: %s", cli_exc)
+            return None
+'''
+
 
 def _replace_once(content: str, before: str, after: str, label: str) -> str:
     """Apply one exact replacement and fail closed when upstream drifts."""
@@ -311,6 +355,21 @@ def patch_loop(source: Path, destination: Path) -> None:
     )
 
 
+def patch_auxiliary(source: Path, destination: Path) -> None:
+    """Let auxiliary tasks reuse the current Codex CLI access token safely."""
+    content = source.read_text(encoding="utf-8")
+    destination.parent.mkdir(parents=True, exist_ok=True)
+    destination.write_text(
+        _replace_once(
+            content,
+            AUXILIARY_TOKEN_BEFORE,
+            AUXILIARY_TOKEN_AFTER,
+            "Codex CLI auxiliary auth",
+        ),
+        encoding="utf-8",
+    )
+
+
 def main() -> int:
     parser = argparse.ArgumentParser(description=__doc__)
     parser.add_argument("provider_source", type=Path)
@@ -323,12 +382,15 @@ def main() -> int:
     parser.add_argument("fallback_destination", type=Path)
     parser.add_argument("loop_source", type=Path)
     parser.add_argument("loop_destination", type=Path)
+    parser.add_argument("auxiliary_source", type=Path)
+    parser.add_argument("auxiliary_destination", type=Path)
     args = parser.parse_args()
     patch_provider(args.provider_source, args.provider_destination)
     patch_session(args.session_source, args.session_destination)
     patch_turn(args.turn_source, args.turn_destination)
     patch_fallback(args.fallback_source, args.fallback_destination)
     patch_loop(args.loop_source, args.loop_destination)
+    patch_auxiliary(args.auxiliary_source, args.auxiliary_destination)
     return 0
 
 

testing/tests/test_hermes_cli_lanes.py

@@ -957,6 +957,70 @@ def test_codex_runtime_patch_uses_cli_and_forwards_route(tmp_path: Path):
     assert 'if agent.api_mode == "codex_app_server"' in loop_content
     assert "build_cross_provider_codex_prompt" in loop_content
 
+    auxiliary = tmp_path / "auxiliary_client.py"
+    auxiliary.write_text(
+        codex_runtime_patch.AUXILIARY_TOKEN_BEFORE,
+        encoding="utf-8",
+    )
+    auxiliary_out = tmp_path / "patched/auxiliary_client.py"
+    codex_runtime_patch.patch_auxiliary(auxiliary, auxiliary_out)
+    auxiliary_content = auxiliary_out.read_text()
+    assert 'os.environ.get("CODEX_HOME"' in auxiliary_content
+    assert 'Path(codex_home).expanduser() / "auth.json"' in auxiliary_content
+    assert "sole owner" in auxiliary_content
+
+
+def test_agent_mounts_codex_auxiliary_runtime_patch():
+    deployment = _agent_deployment()
+    pod = deployment["spec"]["template"]["spec"]
+    patch_init = next(
+        item for item in pod["initContainers"]
+        if item["name"] == "patch-codex-runtime"
+    )
+    assert patch_init["command"][-2:] == [
+        "/opt/hermes/agent/auxiliary_client.py",
+        "/patched/auxiliary_client.py",
+    ]
+    expected_mount = {
+        "name": "codex-runtime-patch",
+        "mountPath": "/opt/hermes/agent/auxiliary_client.py",
+        "subPath": "auxiliary_client.py",
+    }
+    containers = {item["name"]: item for item in pod["containers"]}
+    for name in ("hermes", "terminal"):
+        assert expected_mount in containers[name]["volumeMounts"]
+
+
+def test_codex_auxiliary_patch_reads_cli_token_without_copying_it(
+    tmp_path: Path,
+    monkeypatch,
+):
+    codex_home = tmp_path / ".codex"
+    codex_home.mkdir()
+    (codex_home / "auth.json").write_text(
+        json.dumps({"tokens": {"access_token": "cli-access-token"}}),
+        encoding="utf-8",
+    )
+    monkeypatch.setenv("CODEX_HOME", str(codex_home))
+
+    source = tmp_path / "auxiliary_client.py"
+    source.write_text(
+        "import json, logging, os, time\n"
+        "from pathlib import Path\n"
+        "logger = logging.getLogger(__name__)\n"
+        "def read_token():\n"
+        "    try:\n"
+        "        raise RuntimeError('Hermes provider store intentionally empty')\n"
+        + codex_runtime_patch.AUXILIARY_TOKEN_BEFORE,
+        encoding="utf-8",
+    )
+    destination = tmp_path / "patched/auxiliary_client.py"
+    codex_runtime_patch.patch_auxiliary(source, destination)
+    namespace: dict = {}
+    exec(compile(destination.read_text(), str(destination), "exec"), namespace)
+
+    assert namespace["read_token"]() == "cli-access-token"
+
 
 def test_codex_runtime_patch_fails_closed_on_upstream_drift(tmp_path: Path):
     source = tmp_path / "runtime_provider.py"