codex - 💡(How to fix) Fix codex-cli stalls for exactly 300s between tool result and next model request (stream disconnected)

{
  "schemaVersion": 1,
  "generatedAt": "1779332610s since unix epoch",
  "overallStatus": "fail",
  "codexVersion": "0.132.0",
  "checks": {
    "app_server.status": {
      "id": "app_server.status",
      "category": "app-server",
      "status": "ok",
      "summary": "background server is not running",
      "details": {
        "control socket": "C:\\Users\\zhouj\\.codex\\app-server-control\\app-server-control.sock",
        "daemon state dir": "C:\\Users\\zhouj\\.codex\\app-server-daemon",
        "mode": "ephemeral",
        "pid file": "C:\\Users\\zhouj\\.codex\\app-server-daemon\\app-server.pid (missing)",
        "settings": "C:\\Users\\zhouj\\.codex\\app-server-daemon\\settings.json (missing)",
        "status": "not running",
        "update-loop pid file": "C:\\Users\\zhouj\\.codex\\app-server-daemon\\app-server-updater.pid (missing)"
      },
      "remediation": null,
      "durationMs": 0
    },
    "auth.credentials": {
      "id": "auth.credentials",
      "category": "auth",
      "status": "ok",
      "summary": "auth is configured",
      "details": {
        "auth file": "C:\\Users\\zhouj\\.codex\\auth.json",
        "auth storage mode": "File",
        "stored API key": "true",
        "stored ChatGPT tokens": "false",
        "stored agent identity": "false",
        "stored auth mode": "api_key"
      },
      "remediation": null,
      "durationMs": 0
    },
    "config.load": {
      "id": "config.load",
      "category": "config",
      "status": "ok",
      "summary": "config loaded",
      "details": {
        "CODEX_HOME": "C:\\Users\\zhouj\\.codex",
        "config.toml": "C:\\Users\\zhouj\\.codex\\config.toml",
        "config.toml parse": "ok",
        "enabled feature flags": "shell_tool, shell_snapshot, terminal_resize_reflow, sqlite, hooks, enable_request_compression, multi_agent, apps, tool_search, tool_suggest, plugins, plugin_hooks, in_app_browser, browser_use, browser_use_external, computer_use, plugin_sharing, image_generation, skill_mcp_dependency_install, steer, guardian_approval, goals, collaboration_modes, tool_call_mcp_elicitation, personality, fast_mode, tui_app_server, workspace_dependencies",
        "feature flag overrides": "goals=true",
        "feature flags enabled": "28",
        "log dir": "C:\\Users\\zhouj\\.codex\\log",
        "mcp servers": "0",
        "model": "gpt-5.5",
        "model provider": "custom",
        "sqlite home": "C:\\Users\\zhouj\\.codex"
      },
      "remediation": null,
      "durationMs": 0
    },
    "installation": {
      "id": "installation",
      "category": "install",
      "status": "ok",
      "summary": "installation looks consistent",
      "details": {
        "PATH codex #1": "C:\\Users\\zhouj\\AppData\\Roaming\\npm\\codex",
        "PATH codex #2": "C:\\Users\\zhouj\\AppData\\Roaming\\npm\\codex.cmd",
        "PATH codex entries": "2",
        "current executable": "C:\\Users\\zhouj\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc\\codex\\codex.exe",
        "install context": "npm",
        "managed by bun": "false",
        "managed by npm": "true",
        "managed package root": "C:\\Users\\zhouj\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex",
        "npm update target": "C:\\Users\\zhouj\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex"
      },
      "remediation": null,
      "durationMs": 358
    },
    "mcp.config": {
      "id": "mcp.config",
      "category": "mcp",
      "status": "ok",
      "summary": "no MCP servers configured",
      "details": {},
      "remediation": null,
      "durationMs": 0
    },
    "network.env": {
      "id": "network.env",
      "category": "network",
      "status": "ok",
      "summary": "network-related environment looks readable",
      "details": {
        "proxy env vars present": "NO_PROXY, no_proxy"
      },
      "remediation": null,
      "durationMs": 0
    }
  }
}

---

model_provider = "custom"
   model = "gpt-5.5"
   model_reasoning_effort = "xhigh"
   sandbox_mode = "danger-full-access"
   
   [model_providers.custom]
   name = "custom"
   wire_api = "responses"
   supports_websockets = false
   requires_openai_auth = true
   base_url = "http://127.0.0.1:15721/v1"

2. Start a codex session with a complex task that requires multiple tool calls (file reads, shell commands, code edits)

3. Let it run for 30+ minutes

4. Observe periodic 300-second gaps where no HTTP requests are sent to the model endpoint, followed by "stream disconnected - retrying sampling request (1/5)"

Detailed sequence before first stall:

12:20:40 client: close (model returned tool_call response, time.idle=29.5s)
12:20:40 ToolCall: shell_command (read test_supervisor.py)
12:20:40 ToolCall: shell_command (read test_session_adapter.py)
12:20:40 ToolCall: shell_command (git diff)
12:21:03 client: close (model responded, time.idle=22.4s)
12:21:04 ToolCall: shell_command (Get-Content SKILL.md) <- executes in ~200ms
12:21:04 client: new   <- sends tool result back to model
12:21:44 client: close time.idle=40.2s  <- model responds normally
         *** 300 seconds of silence — no client:new, no ToolCall ***
12:26:44 stream disconnected - retrying (1/5 in 197ms)
12:26:44 client: new   <- retry works instantly
The model's response at 12:21:44 completes normally. After client: close, codex_core should either issue the next client: new or complete the turn. Instead, nothing happens for exactly 300 seconds.

No thread ID available (custom provider endpoint).

### What is the expected behavior?

After `client: close`, the turn should immediately either:
1. Dispatch the next tool call from the response, or
2. Send the next `client: new` request with tool results, or
3. Complete the turn

There should be no 300-second idle gap between a successful response and the next action. The 300s value is suspiciously precise (300.003s and 300.021s), suggesting a hardcoded timeout constant in `codex_core::session::turn` that fires when the turn state machine fails to advance after receiving a complete response.

### Additional information

**Analysis:**

The 300-second value appears to be a hardcoded stream-activity watchdog in `codex_core::session::turn`. The pattern suggests a race condition where the turn state machine sometimes fails to advance after receiving a complete model response. It sits idle until the watchdog fires and forces a retry.

**This is NOT a network issue because:**
- The same local proxy handles Claude Code traffic without any stalls
- The retry after 300s succeeds in <250ms every time
- `time.idle` values on normal requests (22-134s) show the model endpoint is responsive

**Possibly related:**
- `app-server event consumer lagged; dropping ignored events skipped=93/135/231` — TUI event loop under pressure, may indicate backpressure blocking the turn from advancing
- Multiple `codex_core::tools::router: error=Exit code: 1` errors in the same session (tool execution failures)
- `model_reasoning_effort = "xhigh"` produces longer model responses and more tool calls per turn, possibly increasing the chance of hitting the race condition

**Suggested investigation areas:**
1. The 300s constant in `codex_core::session::turn` — is this a stream keepalive/watchdog timer?
2. Possible race condition between the response parser completing and the turn state machine's next-step dispatch
3. Whether `app-server event consumer lagged` indicates backpressure that blocks the turn from advancing
4. Whether the `responses` wire API path has a different state machine than the `chat` path

**Log grep patterns for reproduction:**