New OpenAI harness Stuff

[hybrid jackal]

I updated to 2026.5.12 last night to take advantage of the new openai harness, being that I've been using oauth with anthropic api as a backup. My channel is through discord. Whenever I try to communicate with my bot while the model is set to gpt 5.5, it just hangs while typing for minutes. It then either just lets go and times out, or swaps to the fallback to respond to me.

Checking logs now and realizing that I may just have to reauth, but I already did that and got back to this issue. I can try it again.
Then there are a lot of stalled session logs. Walk me through how to fix this please.

[sinful burrow]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[bronze marsh]

Josh, I’d treat this as routing/session-state first, not “reauth again” first.

Run this order:

openclaw --version
openclaw status --all
openclaw models status
openclaw models auth list --provider openai-codex
openclaw config get agents.defaults.model --json
openclaw config get models.providers.openai.agentRuntime --json

What you want to see: the agent model route should be canonical openai/gpt-5.5, while Codex subscription auth can still show up as openai-codex:.... The old openai-codex/gpt-* model refs are legacy and can leave stale route/session pins around.

Then run the repair pass:

[bronze marsh]

openclaw doctor --fix
openclaw config validate
openclaw gateway restart --safe

If the restart refuses because the gateway thinks work is still active, that matches your “stalled session” logs. Inspect the status output first, then as the operator escape hatch you can do:

openclaw gateway restart --safe --skip-deferral

After restart, in the affected Discord session/thread, start fresh with /reset or /new so it doesn’t keep a stale model/auth/runtime pin.

Only reauth if this shows no usable Codex profile:

[bronze marsh]

openclaw models auth login --provider openai-codex
# headless/VPS:
openclaw models auth login --provider openai-codex --device-code
openclaw models status --probe --probe-provider openai-codex

Tiny warning: --probe makes real provider calls, so use it intentionally.

Also: don’t chase channels.discord.eventQueue.listenerTimeout unless the logs specifically show Discord listener timeouts. The “stuck session / processing” style logs usually mean the agent turn/runtime is hung or pinned, not that Discord itself needs a bigger timeout.

Docs I checked:
https://docs.openclaw.ai/help/faq
https://docs.openclaw.ai/providers/openai
https://docs.openclaw.ai/concepts/models
https://docs.openclaw.ai/concepts/model-failover

[bronze marsh]

If it still hangs after that, grab a sanitized bundle and file it:

openclaw gateway diagnostics export --output openclaw-diagnostics.zip

Issues: https://github.com/openclaw/openclaw/issues

[hybrid jackal]

agent responded to me once when i did /new but now its hanging again. so i ran the oauth setup again, got this warning though.
[agents/auth-profiles] failed to create oauth profile secret keychain entry

am I doing something wrong probably?

[bronze marsh]

Probably not you. That warning means OpenClaw tried to create the macOS Keychain master entry used to encrypt OAuth profile secret material and Keychain rejected it. It should then fall back to a local key file outside the state dir, so the warning alone is not necessarily fatal.

Check whether the login actually persisted:

openclaw models auth list --provider openai-codex
openclaw models status
openclaw models status --probe --probe-provider openai-codex

If the profile shows up and the probe is OK, stop reauthing. The hang is likely the Codex runtime/session path, not missing OAuth.

If the profile does not show up, or probe says missing/expired/unresolved, then the secret fallback may also be failing. On macOS check whether one of these exists after login:

ls -l "$HOME/Library/Application Support/OpenClaw/auth-profile-secret-key" 2>/dev/null

[bronze marsh]

ls -l "$HOME/.openclaw-auth-profile-secrets/auth-profile-secret-key" 2>/dev/null

Those are not OAuth tokens, but still don’t paste their contents anywhere.

Also grab the actual error next to that warning:

openclaw logs --limit 300 --plain | grep -i -A3 -B3 "oauth profile secret keychain"

Common causes are: macOS Keychain locked/not available to the user running the command, OpenClaw running under a different user/session than the one doing OAuth, or permissions problems creating the fallback key file.

Given /new worked once and then it hung again, I’d also do this after the auth check:

[bronze marsh]

openclaw doctor --fix
openclaw gateway restart --safe --skip-deferral

Then start another fresh Discord session with /new.

Docs context:
https://docs.openclaw.ai/concepts/oauth
https://docs.openclaw.ai/providers/openai
https://docs.openclaw.ai/cli/models

If the probe passes but GPT-5.5 still hangs, that’s issue-worthy with logs/diagnostics because auth is probably not the root cause:

openclaw gateway diagnostics export --output openclaw-diagnostics.zip

https://github.com/openclaw/openclaw/issues

[hybrid jackal]

tried everythign it didn't work, and / new just takes forever to get any sort of response back if I even get one. help me get the pasteable logs that will be relevant to you in helping me fix this please.

[bronze marsh]

Yep. Let’s collect a tight bundle of text that is useful here without dumping secrets or 20k lines.

Run this first and paste the output:

openclaw --version
openclaw status --all
openclaw models status
openclaw models auth list --provider openai-codex
openclaw config get agents.defaults.model --json
openclaw config get models.providers.openai.agentRuntime --json

Then reproduce the hang once:

1. In Discord, send /new
2. Wait until it clearly stalls or falls back
3. Immediately run this:

[bronze marsh]

openclaw logs --limit 1500 --plain |
  rg -i -C 4 "openai|codex|gpt-5\.5|auth-profiles|oauth|keychain|stuck session|processing|timeout|fallback|cooldown|rate_limit|subscription|app-server|runtime|error|warn"

If rg is not installed, use:

openclaw logs --limit 1500 --plain |
  grep -i -E -C 4 "openai|codex|gpt-5\.5|auth-profiles|oauth|keychain|stuck session|processing|timeout|fallback|cooldown|rate_limit|subscription|app-server|runtime|error|warn"

Before pasting, quickly scan for anything that looks like a token/key/cookie. openclaw status --all and openclaw logs should redact secrets, but still eyeball it.

If the filtered log is still huge, paste:

• the first chunk around the first OpenAI/Codex error
• the chunk around stuck session / timeout

[bronze marsh]

• the final 80-ish lines

Also run this locally but don’t paste the zip contents here:

openclaw gateway diagnostics export --output openclaw-diagnostics.zip

If we can’t solve it from the pasted text, that zip is the thing to attach to a GitHub issue: https://github.com/openclaw/openclaw/issues

[hybrid jackal]

026-05-16T03:24:10.941Z info channels/discord {"subsystem":"channels/discord"} discord channels resolved: 1494084111164506177/1494084112875524190 (guild:Gravity Chamber; channel:main), 1494084111164506177/1494084302202474587 (guild:Gravity Chamber; channel:accountant), 1494084111164506177/1494084379817938944 (guild:Gravity Chamber; channel:notes), 1494084111164506177/1494084466480644271 (guild:Gravity Chamber; channel:health), 1494084111164506177/1494084512840286510 (guild:Gravity Chamber; channel:work), 1494084111164506177/1494084578506309662 (guild:Gravity Chamber; channel:calendar) (+4)
2026-05-16T03:24:11.198Z info channels/discord {"subsystem":"channels/discord"} discord client initialized as 1494090736965124196; awaiting gateway readiness
2026-05-16T03:24:42.295Z warn plugins {"subsystem":"plugins"} [plugins] plugins.allow is empty; discovered non-bundled plugins may auto-load: codex (/Users/server/.openclaw/npm/node_modules/@openclaw/codex/dist/index.js), discord (/Users/server/.openclaw/npm/node_modules/@openclaw/discord/dist/index.js). Set plugins.allow to explicit trusted ids.
2026-05-16T03:25:11.896Z warn diagnostic {"subsystem":"diagnostic"} liveness warning: reasons=event_loop_delay interval=37s eventLoopDelayP99Ms=24.7 eventLoopDelayMaxMs=29293 eventLoopUtilization=0.795 cpuCoreRatio=0.573 active=1 waiting=0 queued=1 phase=channels.telegram.start-account recentPhases=post-attach.update-sentinel:20ms,sidecars.session-locks:35ms,sidecars.model-prewarm:678ms,post-attach.tailscale:757ms,gateway.ready:4526ms,post-ready.maintenance:24ms work=[active=agent:main:discord:slash:144267499788042240(processing,q=1,age=30s) queued=agent:main:discord:slash:144267499788042240(processing,q=1,age=30s)]
2026-05-16T03:25:23.847Z warn fetch-timeout {"subsystem":"fetch-timeout"} {"timeoutMs":10000,"elapsedMs":11565,"operation":"fetchWithTimeout","url":"https://discord.com/api/v10/oauth2/applications/@me"} fetch timeout reached; aborting operation

2026-05-16T03:24:09.561Z info channels/discord {"subsystem":"channels/discord"} [default] Discord bot probe resolved @Roshi
2026-05-16T03:24:09.564Z info channels/discord {"subsystem":"channels/discord"} [default] Discord Message Content Intent is limited; bots under 100 servers can use it without verification.

[bronze marsh]

This snippet is useful. Two things jump out:

1. The gateway event loop is getting wedged for ~29s:
   eventLoopDelayMaxMs=29293 ... active=1 queued=1 phase=channels.telegram.start-account

That means your Discord slash turn is queued/processing while the gateway is busy starting Telegram. This does not look like “OAuth missing” from this snippet.

2. Discord metadata lookup is timing out:
   discord.com/api/v10/oauth2/applications/@me

That can be network slowness, but the event-loop warning is the bigger clue.

For isolation, temporarily turn Telegram off and restart, then try Discord /new again:

openclaw config set channels.telegram.enabled false

[bronze marsh]

openclaw config validate
openclaw gateway restart --force

If Discord becomes responsive after that, the problem is in Telegram startup/network pressure, not GPT-5.5 auth. You can turn it back on later with:

openclaw config set channels.telegram.enabled true
openclaw gateway restart --force

Next logs I need after trying with Telegram disabled:

openclaw logs --limit 800 --plain |

[bronze marsh]

  rg -i -C 5 "event_loop_delay|channels.telegram|channels.discord|slash|stuck session|codex|openai|gpt-5\.5|app-server|timeout|fallback|error|warn"

Also paste these if you haven’t yet:

openclaw --version
openclaw status --all
openclaw models status
openclaw models auth list --provider openai-codex

The plugins.allow is empty warning is worth cleaning up later, but I would not chase it as the immediate hang unless you’re loading unexpected plugins. The active symptom here is gateway liveness/event-loop delay during Telegram account startup.

[bronze marsh]

Those two Discord lines are normal/info-level. They mean the bot token resolved correctly and Discord is reminding you about Message Content Intent limits. They don’t explain the hang.

The important line is still the liveness one:

eventLoopDelayMaxMs=29293 ... phase=channels.telegram.start-account ... active=agent:main:discord:slash...

So I’d keep going with the Telegram isolation test from above. After disabling Telegram + restarting, try one Discord /new, then paste the filtered log output. That will tell us whether Telegram startup is blocking the gateway or whether Codex/OpenAI is the next failure after the gateway is responsive.

[hybrid jackal]

dont have rg so i cant get those logs the way you want them, and no nitro so i cant paste everytthing i want to in here. but i isolated it and still i can't do anything

[bronze marsh]

Got it. Don’t fight Discord paste limits. Use built-in tools only and make a small text file you can upload here.

Run this right after one failed /new:

openclaw --version > openclaw-debug.txt
printf '
===== STATUS =====
' >> openclaw-debug.txt
openclaw status --all >> openclaw-debug.txt 2>&1
printf '
===== MODELS =====
' >> openclaw-debug.txt
openclaw models status >> openclaw-debug.txt 2>&1
printf '
===== OPENAI CODEX AUTH =====

[bronze marsh]

' >> openclaw-debug.txt
openclaw models auth list --provider openai-codex >> openclaw-debug.txt 2>&1
printf '
===== MODEL CONFIG =====
' >> openclaw-debug.txt
openclaw config get agents.defaults.model --json >> openclaw-debug.txt 2>&1
openclaw config get models.providers.openai.agentRuntime --json >> openclaw-debug.txt 2>&1
printf '
===== FILTERED LOGS =====
' >> openclaw-debug.txt
openclaw logs --limit 1200 --plain 2>&1 | grep -i -E -C 5 "event_loop_delay|channels.telegram|channels.discord|slash|stuck session|processing|codex|openai|gpt-5\.5|app-server|auth-profiles|oauth|keychain|timeout|fallback|cooldown|rate_limit|subscription|error|warn" >> openclaw-debug.txt

Then upload openclaw-debug.txt as a file attachment here. A .txt attachment should avoid the Nitro message-length wall.

If that file is still too large, make a smaller one:

[bronze marsh]

tail -n 250 openclaw-debug.txt > openclaw-debug-tail.txt

and upload openclaw-debug-tail.txt.

Since you isolated Telegram and it still fails, the next likely bucket is Codex app-server/runtime/auth routing or a stuck Discord slash session, so the status + model/auth + filtered logs should be enough to narrow it.

[hybrid jackal]

dude this channel doesnt let me upload files

[bronze marsh]

Fair. Let’s go tiny and paste in chunks here. No attachments, no rg.

Run/paste these one at a time. Each should be short.

1. Version + health

openclaw --version
openclaw status --all | tail -n 80

2. Model/auth routing

openclaw models status | tail -n 80
openclaw models auth list --provider openai-codex
openclaw config get agents.defaults.model --json
openclaw config get models.providers.openai.agentRuntime --json

[bronze marsh]

3. Only recent warnings/errors

openclaw logs --limit 250 --plain | grep -i -E "warn|error|timeout|stuck|event_loop|codex|openai|oauth|auth-profiles|fallback" | tail -n 80

4. If that is still too big, use this smaller one

openclaw logs --limit 120 --plain | tail -n 80

Paste chunk 1 first, then chunk 2. We don’t need everything at once. I’m mainly looking for: the exact version, whether openai/gpt-5.5 is routed through Codex, whether the Codex auth profile is usable, and the first real warning/error after the failed /new.

[hybrid jackal]

its all still way too big whenever i try to copy paste the status stuff over. atp just give me a summary so i can talk to chatgpt and send it the logs cause this is too much of a hassle

[bronze marsh]

Yeah, that’s frustrating. Here’s a clean summary you can give ChatGPT or attach to an issue.

OpenClaw Discord bot hangs after upgrading to 2026.5.12 and switching to the new OpenAI/Codex harness with model openai/gpt-5.5.

Setup:
- Channel: Discord
- Model target: GPT-5.5 / OpenAI Codex harness
- Auth: OpenAI Codex OAuth, with Anthropic fallback available
- Symptom: Discord bot starts typing for minutes, often times out or falls back. /new responded once, then hangs again.
- Re-running OAuth produced warning:
  [agents/auth-profiles] failed to create oauth profile secret keychain entry

Things already tried:
- Re-ran OpenAI Codex OAuth
- Ran doctor/repair guidance

[bronze marsh]

- Used /new and /reset style fresh session attempts
- Isolated/disabled Telegram as a suspected startup blocker, but Discord still hangs

Important log snippets:
- Discord starts and resolves channels:
  channels/discord: discord channels resolved ...
  channels/discord: discord client initialized ... awaiting gateway readiness
  channels/discord: Discord bot probe resolved @Roshi
- Plugin warning:
  [plugins] plugins.allow is empty; discovered non-bundled plugins may auto-load: codex (.../@openclaw/codex/dist/index.js), discord (.../@openclaw/discord/dist/index.js)
- Gateway liveness warning:
  liveness warning: reasons=event_loop_delay interval=37s eventLoopDelayP99Ms=24.7 eventLoopDelayMaxMs=29293 eventLoopUtilization=0.795 cpuCoreRatio=0.573 active=1 waiting=0 queued=1 phase=channels.telegram.start-account work=[active=agent:main:discord:slash:<user>(processing,q=1,age=30s) queued=agent:main:discord:slash:<user>(processing,q=1,age=30s)]
- Discord HTTP timeout:
  fetch-timeout timeoutMs=10000 elapsedMs=11565 url=https://discord.com/api/v10/oauth2/applications/@me

[bronze marsh]

Current suspicion:
- The keychain warning may not be fatal by itself because OpenClaw should fall back to a local auth-profile secret key file, but it may indicate auth persistence problems if fallback also failed.
- The event_loop_delay line shows the gateway event loop was badly delayed while a Discord slash session was processing/queued. Initially this looked related to Telegram startup, but disabling Telegram did not fix it.
- The remaining likely areas are:
  1. Codex app-server/runtime hang
  2. stale or stuck Discord slash session state
  3. OpenAI Codex OAuth profile exists but is unusable/expired/unreadable
  4. gateway event loop starvation after startup
  5. plugin/runtime issue around @openclaw/codex in 2026.5.12

Useful commands to ask about:
- openclaw --version
- openclaw status --all
- openclaw models status
- openclaw models auth list --provider openai-codex