update is causing error

[chrome leaf]

@old stream
i just updated openclaw and get this error now
Agent failed before reply: Requested agent harness "codex" is not registered.
Logs: openclaw logs --follow

[echo dust]

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

[old stream]

That error usually means your config is forcing the embedded harness to codex, but the bundled codex plugin did not load/register after the update.

First, can you paste:

openclaw --version
openclaw plugins list --json
openclaw config get agents.defaults.embeddedHarness --json
openclaw config get plugins --json

Quick unblock if you did not intentionally switch to the native Codex harness:

openclaw config set agents.defaults.embeddedHarness.runtime '"auto"' --strict-json
openclaw gateway restart

[old stream]

If you did intend to use the Codex harness, make sure the bundled plugin is enabled and not excluded by plugins.allow:

openclaw config set plugins.entries.codex.enabled true --strict-json
openclaw gateway restart

Docs: https://docs.openclaw.ai/plugins/codex-harness and https://docs.openclaw.ai/gateway/configuration-reference#agentsdefaultsembeddedharness

[chrome leaf]

what is the differences between codex harness? we had to use our codex subscription initially to use gpt at one time, could it have been from that configuration?

[old stream]

Yes, very likely.

There are two different “Codex” paths that sound similar:

1. Codex subscription / OAuth as a model provider

   • Config uses model refs like openai-codex/...
   • OpenClaw still runs the normal embedded agent runner
   • This was the path people used when they wanted GPT access through their Codex/ChatGPT subscription auth

2. Codex harness

   • Config forces agents.defaults.embeddedHarness.runtime = "codex"
   • OpenClaw hands the low-level agent turn to the Codex app-server harness
   • This requires the bundled codex plugin to load and register the harness

Your error specifically says the harness named codex is requested but not registered, so it points at path #2, not just “using Codex OAuth”.

So yes: an older config from when you were trying to use GPT via Codex could have left behind one of these:

[old stream]

agents: {
  defaults: {
    embeddedHarness: {
      runtime: "codex"
    }
  }
}

or an env var like:

OPENCLAW_AGENT_RUNTIME=codex

If you only need the subscription/OAuth provider path, set the harness back to auto:

[old stream]

openclaw config set agents.defaults.embeddedHarness.runtime '"auto"' --strict-json
openclaw gateway restart

Docs for the split: https://docs.openclaw.ai/plugins/codex-harness

[chrome leaf]

i tried those commands, still having same error

[old stream]

Two likely reasons it can persist after that:

1. The existing thread/session is pinned to the old harness. Harness choice is sticky per session. Try starting a fresh session in this thread:

/reset

or /new, then send a normal test message.

2. Something else is still forcing codex, usually an env var or per-agent override. Please paste these outputs:

openclaw --version
openclaw config get agents.defaults.embeddedHarness --json
openclaw config get agents.list --json

[old stream]

openclaw config get plugins --json
printenv | grep OPENCLAW_AGENT
openclaw logs --tail 120

Also check whether you set this in your service env, shell profile, Docker compose, systemd unit, or launchd plist:

OPENCLAW_AGENT_RUNTIME=codex

If that env var exists, it overrides config. Remove it or set it to auto, then restart the gateway/service.

[chrome leaf]

openclaw logs --tail 120
{
"entries": {
"telegram": {
"enabled": true,
"config": {}
},
"discord": {
"enabled": true,
"config": {}
},
"brave": {
"enabled": true,
"config": {
"webSearch": {
"apiKey": "OPENCLAW_REDACTED"
}
}
},
"openai": {
"enabled": true,
"config": {
"personality": "friendly"
}
},
"anthropic": {
"enabled": true,
"config": {}
},
"github-copilot": {
"enabled": true,
"config": {}
},
"memory-core": {
"enabled": true,
"config": {
"dreaming": {
"enabled": true
}
}
}
}
}
OpenClaw does not recognize option "--tail".
Try: openclaw logs 120 --help

[old stream]

My bad on the logs flag. Current docs say it is:

openclaw logs --limit 120

That plugin output is useful: I do not see codex enabled there. So if OpenClaw is still asking for harness codex, it’s almost certainly being forced by either:

• agents.defaults.embeddedHarness.runtime
• an agents.list[*].embeddedHarness.runtime per-agent override
• OPENCLAW_AGENT_RUNTIME=codex
• or the existing session is still pinned and needs /reset

Try the stronger config cleanup:

[old stream]

openclaw config unset agents.defaults.embeddedHarness
openclaw gateway restart

Then in this Discord thread run:

/reset

If it still fails, please paste these exact outputs:

openclaw --version
openclaw config get agents.defaults.embeddedHarness --json

[old stream]

openclaw config get agents.list --json
printenv | grep OPENCLAW_AGENT
openclaw logs --limit 120

If agents.list shows an agent with "embeddedHarness": { "runtime": "codex" }, that override needs to be removed too. Docs: https://docs.openclaw.ai/plugins/codex-harness and https://docs.openclaw.ai/cli/logs

[chrome leaf]

first part gave me this, and still having the same initial error...
bash openclaw config unset agents.defaults.embeddedHarness
openclaw gateway restart
/opt/homebrew/bin/openclaw: line 3: import: command not found
/opt/homebrew/bin/openclaw: line 4: import: command not found
/opt/homebrew/bin/openclaw: line 5: import: command not found
/opt/homebrew/bin/openclaw: line 6: import: command not found
/opt/homebrew/bin/openclaw: line 7: import: command not found
/opt/homebrew/bin/openclaw: line 8: import: command not found
/opt/homebrew/bin/openclaw: line 9: import: command not found
/opt/homebrew/bin/openclaw: line 11: const: command not found
/opt/homebrew/bin/openclaw: line 12: const: command not found
/opt/homebrew/bin/openclaw: line 13: .: filename argument required
.: usage: . filename [arguments]
/opt/homebrew/bin/openclaw: line 13: const: command not found
/opt/homebrew/bin/openclaw: openclaw: line 15: syntax error near unexpected token (' /opt/homebrew/bin/openclaw: openclaw: line 15: const parseNodeVersion = (rawVersion) => {'
Config warnings:

• plugins.entries.discord: plugin not installed: discord — install the official external plugin with: openclaw plugins install @openclaw/discord
• plugins.entries.brave: plugin not installed: brave — install the official external plugin with: openclaw plugins install @openclaw/brave-plugin

The second part command returned alot of text, i may have to paste multiple posts for you to see it all

[old stream]

Ah, I see the issue: you accidentally ran the OpenClaw CLI through bash.

This part:

bash openclaw config unset agents.defaults.embeddedHarness

makes bash try to interpret the OpenClaw JavaScript file, which is why you got import: command not found and const: command not found. The config unset probably did not happen.

Run it exactly like this, no bash in front:

openclaw config unset agents.defaults.embeddedHarness
openclaw config get agents.defaults.embeddedHarness --json
openclaw gateway restart

[old stream]

Then in this Discord thread run:

/reset

If config get agents.defaults.embeddedHarness still shows anything with "runtime": "codex", paste that small output.

The plugin warnings are a separate clue too:

plugins.entries.discord: plugin not installed
plugins.entries.brave: plugin not installed

That means your config references external plugins that are no longer installed after the update. You’ll probably want to reinstall those after this harness issue is fixed:

[old stream]

openclaw plugins install @openclaw/discord
openclaw plugins install @openclaw/brave-plugin
openclaw gateway restart

But first fix the stuck harness config/session, because that matches the original Requested agent harness "codex" is not registered error.

[chrome leaf]

/reset

ok i typed the reset command in this chat then did config get agents.defaults.embeddedHarness and got this:
Config warnings:

• plugins.entries.discord: plugin not installed: discord — install the official external plugin with: openclaw plugins install @openclaw/discord
• plugins.entries.brave: plugin not installed: brave — install the official external plugin with: openclaw plugins install @openclaw/brave-plugin

🦞 OpenClaw 2026.5.12 (f066dd2) — iMessage green bubble energy, but for everyone.

Config path not found: agents.defaults.embeddedHarness. Run openclaw config validate to inspect config shape.

[old stream]

Good, Config path not found: agents.defaults.embeddedHarness means the default harness override is gone. So the remaining source is probably one of these:

• per-agent override in agents.list
• service/env override like OPENCLAW_AGENT_RUNTIME=codex
• plugin install state after the update

Next, run these smaller checks:

openclaw config get agents.list --json | grep -n -C 3 'embeddedHarness\|codex'
printenv | grep OPENCLAW_AGENT
launchctl getenv OPENCLAW_AGENT_RUNTIME
openclaw config validate

On macOS, printenv only checks your current shell. If the gateway is running under launchd, launchctl getenv OPENCLAW_AGENT_RUNTIME matters more.

[old stream]

Also, those plugin warnings are real. Your config says Discord and Brave are enabled, but they are not installed in this OpenClaw install. Reinstall them, then restart:

openclaw plugins install @openclaw/discord
openclaw plugins install @openclaw/brave-plugin
openclaw gateway restart

After restart, run /reset again in the Discord thread and test. If the harness error still appears, paste just the output of the agents.list ... grep and the two env checks above.

[chrome leaf]

first command:
max@12:74:81:ad:54:8b ~ % openclaw config get agents.list --json | grep -n -C 3 'embeddedHarness|codex'
printenv | grep OPENCLAW_AGENT
launchctl getenv OPENCLAW_AGENT_RUNTIME
openclaw config validate
7- "name": "bdcbot",
8- "workspace": "/Users/max/.openclaw/workspace",
9- "agentDir": "/Users/max/.openclaw/agents/bdcbot/agent",
10: "model": "openai-codex/gpt-5.4"
11- }
12-]
Config warnings:

• plugins.entries.discord: plugin not installed: discord — install the official external plugin with: openclaw plugins install @openclaw/discord
• plugins.entries.brave: plugin not installed: brave — install the official external plugin with: openclaw plugins install @openclaw/brave-plugin

🦞 OpenClaw 2026.5.12 (f066dd2) — Gateway online—please keep hands, feet, and appendages inside the shell at all times.

Config valid: ~/.openclaw/openclaw.json

[old stream]

That output narrows it down nicely.

openai-codex/gpt-5.4 is your old Codex subscription/OAuth model route. By itself, that should be the PI/OpenClaw provider path, not the native codex harness. Since there’s no default embeddedHarness and no env output, I’d fix the install/plugin state next.

Your config says Discord and Brave are enabled, but this install says those plugins are missing. Run:

openclaw plugins install @openclaw/discord
openclaw plugins install @openclaw/brave-plugin
openclaw gateway restart

Then in this Discord thread:

/reset

[old stream]

If it still throws the same harness error, don’t paste the huge full logs yet. Paste this filtered output instead:

openclaw logs --limit 300 --plain | grep -i -C 4 'harness\|codex\|plugin'
openclaw sessions --all-agents --json | grep -n -C 4 'harness\|codex'

At this point the remaining suspicious thing is a stale session/harness pin or a plugin load failure during startup, and those two filtered commands should show which one it is.

[chrome leaf]

im over the limits of this message by 63000 if i paste the response to that command. anything more specific?

[old stream]

Yep, let’s make it tiny. Run these and paste only the output:

openclaw logs --limit 80 --plain | grep -i 'Requested agent harness\|not registered\|agent harness selected\|loaded plugin\|failed.*plugin'

And this config/session search:

grep -RIn '"runtime": "codex"\|"runtime":"codex"\|embeddedHarness\|selectedHarnessId\|harness' ~/.openclaw/openclaw.json ~/.openclaw/agents 2>/dev/null | head -80

If that second one is still too big, split it:

grep -RIn '"runtime": "codex"\|"runtime":"codex"\|embeddedHarness' ~/.openclaw/openclaw.json ~/.openclaw/agents 2>/dev/null | head -40

[old stream]

grep -RIn 'selectedHarnessId\|harness' ~/.openclaw/agents/*/sessions* ~/.openclaw/agents/*/agent 2>/dev/null | head -40

I’m specifically looking for either:

• a leftover config/session pin saying runtime: codex or selectedHarnessId: codex
• a startup/plugin line explaining why the codex harness/plugin didn’t register

[chrome leaf]

'Requested agent harness|not registered|agent harness selected|loaded plugin|failed.*plugin'
2026-05-15T19:38:46.515Z error diagnostic {"subsystem":"diagnostic"} lane task error: lane=main durationMs=82 error="Error: Requested agent harness "codex" is not registered."
2026-05-15T19:38:46.516Z error diagnostic {"subsystem":"diagnostic"} lane task error: lane=session:agent:main:main durationMs=83 error="Error: Requested agent harness "codex" is not registered."
2026-05-15T19:38:46.526Z warn model-fallback/decision {"subsystem":"model-fallback/decision"} {"event":"model_fallback_decision","tags":["error_handling","model_fallback","candidate_failed"],"runId":"1a0066eb-6ba5-4e91-a0a0-7be8a7827828","sessionId":"531cee5b-b047-4ae4-88c4-244a607a184f","lane":"main","decision":"candidate_failed","requestedProvider":"openai-codex","requestedModel":"gpt-5.5","candidateProvider":"openai-codex","candidateModel":"gpt-5.5","attempt":1,"total":2,"errorPreview":"Requested agent harness "codex" is not registered.","errorHash":"sha256:7e02b9d21701","fallbackStepType":"fallback_step","fallbackStepFromModel":"openai-codex/gpt-5.5","fallbackStepToModel":"openai-codex/gpt-5.4","fallbackStepFromFailureDetail":"Requested agent harness "codex" is not registered.","fallbackStepChainPosition":1,"fallbackStepFinalOutcome":"next_fallback","nextCandidateProvider":"openai-codex","nextCandidateModel":"gpt-5.4","isPrimary":true,"requestedModelMatched":true,"fallbackConfigured":true} model fallback decision
2026-05-15T19:38:46.599Z error diagnostic {"subsystem":"diagnostic"} lane task error: lane=main durationMs=63 error="Error: Requested agent harness "codex" is not registered."
2026-05-15T19:38:46.600Z error diagnostic {"subsystem":"diagnostic"} lane task error: lane=session:agent:main:main durationMs=64 error="Error: Requested agent harness "codex" is not registered."

2026-05-15T19:38:46.610Z error Embedded agent failed before reply: Requested agent harness "codex" is not registered.
2026-05-15T19:44:26.766Z error diagnostic {"subsystem":"diagnostic"} lane task error: lane=main durationMs=93 error="Error: Requested agent harness "codex" is not registered."
2026-05-15T19:44:26.767Z error diagnostic {"subsystem":"diagnostic"} lane task error: lane=session:agent:main:main durationMs=95 error="Error: Requested agent harness "codex" is not registered."
Log tail truncated (increase --max-bytes).
2026-05-15T19:44:26.778Z warn model-fallback/decision {"subsystem":"model-fallback/decision"} {"event":"model_fallback_decision","tags":["error_handling","model_fallback","candidate_failed"],"runId":"a1cd08b4-c09f-401d-9c21-3e908aabdb47","sessionId":"531cee5b-b047-4ae4-88c4-244a607a184f","lane":"main","decision":"candidate_failed","requestedProvider":"openai-codex","requestedModel":"gpt-5.5","candidateProvider":"openai-codex","candidateModel":"gpt-5.5","attempt":1,"total":2,"errorPreview":"Requested agent harness "codex" is not registered.","errorHash":"sha256:7e02b9d21701","fallbackStepType":"fallback_step","fallbackStepFromModel":"openai-codex/gpt-5.5","fallbackStepToModel":"openai-codex/gpt-5.4","fallbackStepFromFailureDetail":"Requested agent harness "codex" is not registered.","fallbackStepChainPosition":1,"fallbackStepFinalOutcome":"next_fallback","nextCandidateProvider":"openai-codex","nextCandidateModel":"gpt-5.4","isPrimary":true,"requestedModelMatched":true,"fallbackConfigured":true} model fallback decision
2026-05-15T19:44:26.867Z error diagnostic {"subsystem":"diagnostic"} lane task error: lane=main durationMs=80 error="Error: Requested agent harness "codex" is not registered."
2026-05-15T19:44:26.868Z error diagnostic {"subsystem":"diagnostic"} lane task error: lane=session:agent:main:main durationMs=81 error="Error: Requested agent harness "codex" is not registered."

2026-05-15T19:44:26.876Z error Embedded agent failed before reply: Requested agent harness "codex" is not registered.
max@12:74:81:ad:54:8b ~ %

[old stream]

That log is enough. The failure is happening while trying openai-codex/gpt-5.5, then falling back to openai-codex/gpt-5.4. So your Codex OAuth model config is the trigger.

Since you want the Codex OAuth provider path, force the embedded runner to PI explicitly instead of auto:

openclaw config set agents.defaults.embeddedHarness '{"runtime":"pi","fallback":"pi"}' --strict-json
openclaw gateway restart

Then in this thread:

/reset

That should make openai-codex/gpt-5.4 run through the normal PI/OpenClaw route instead of trying to resolve a native codex harness.

[old stream]

If it still fails after that, paste only:

openclaw config get agents.defaults.model --json
openclaw config get agents.list --json | grep -n -C 4 'model\|embeddedHarness'
openclaw config get agents.defaults.embeddedHarness --json

The plugin warnings for Discord/Brave still need cleanup, but they’re separate from this harness error.

[old stream]

Yep, same signature. Those timestamps are before the PI-forcing command I just sent, so the next useful test is:

openclaw config set agents.defaults.embeddedHarness '{"runtime":"pi","fallback":"pi"}' --strict-json
openclaw gateway restart

Then run /reset here and send a test message.

If it still errors after that, grab the new lines only:

openclaw config get agents.defaults.embeddedHarness --json
openclaw logs --limit 80 --plain | grep -i 'Requested agent harness\|agent harness selected\|embedded agent failed'

I’m expecting the config output to show "runtime": "pi". If it does and the error still asks for codex, that’s a bug or a stale session store entry rather than normal config behavior.

[old stream]

That last line is still from 19:44:26, so it’s the same pre-fix failure.

Please run the PI-forcing command now, then restart and test:

openclaw config set agents.defaults.embeddedHarness '{"runtime":"pi","fallback":"pi"}' --strict-json
openclaw config get agents.defaults.embeddedHarness --json
openclaw gateway restart

Then run /reset here and send a test message.

If it fails again, paste only the fresh output from:

openclaw logs --limit 40 --plain | grep -i 'Requested agent harness\|embedded agent failed\|agent harness selected'

[chrome leaf]

here are responses for both:
OpenClaw 2026.5.12 (f066dd2) — It's not "failing," it's "discovering new ways to configure the same thing wrong."

Error: Config validation failed: agents.defaults.embeddedHarness: Unrecognized key: "fallback"

🦞 OpenClaw 2026.5.12 (f066dd2) — Your task has been queued; your dignity has been deprecated.

openclaw config get agents.defaults.model --json
openclaw config get agents.list --json | grep -n -C 4 'model|embeddedHarness'
openclaw config get agents.defaults.embeddedHarness --json
{
"primary": "openai-codex/gpt-5.5",
"fallbacks": [
"openai-codex/gpt-5.4"
]
}
6- "id": "bdcbot",
7- "name": "bdcbot",
8- "workspace": "/Users/max/.openclaw/workspace",
9- "agentDir": "/Users/max/.openclaw/agents/bdcbot/agent",
10: "model": "openai-codex/gpt-5.4"
11- }
12-]
Config path not found: agents.defaults.embeddedHarness. Run openclaw config validate to inspect config shape.
max@12:74:81:ad:54:8b ~ %

[old stream]

Good catch. I gave you a config shape from newer docs; your 2026.5.12 schema only accepts embeddedHarness.runtime, not fallback. More importantly, in this version openai-codex/... model refs default to the codex runtime unless you override the model runtime policy.

For the immediate unblock, don’t use embeddedHarness. Force those legacy openai-codex refs through PI at the model-entry level:

openclaw config unset agents.defaults.embeddedHarness

openclaw config set agents.defaults.models '{"openai-codex/gpt-5.5":{"agentRuntime":{"id":"pi"}},"openai-codex/gpt-5.4":{"agentRuntime":{"id":"pi"}}}' --strict-json --merge

openclaw config get agents.defaults.models --json | grep -n -C 4 'openai-codex\|agentRuntime'

openclaw gateway restart

Then run /reset here and test again.

Why this should work: your logs show the run is selecting openai-codex/gpt-5.5 and openai-codex/gpt-5.4, and 2026.5.12 treats that provider as Codex-runtime by default. Adding agentRuntime.id: "pi" tells OpenClaw to use the old PI/OpenClaw route for those model refs instead of looking for the native codex harness.

[old stream]

Longer-term, the newer/current config direction is different: use openai/gpt-* model refs plus the Codex plugin for subscription-backed Codex runtime. But let’s get you unstuck first.

[chrome leaf]

🦞 OpenClaw 2026.5.12 (f066dd2) — I'm the reason your shell history looks like a hacker-movie montage.

Config path not found: agents.defaults.embeddedHarness. Nothing was changed. Run openclaw config get <path> first if you are unsure of the path.

🦞 OpenClaw 2026.5.12 (f066dd2) — I'm not AI-powered, I'm AI-possessed. Big difference.

Config overwrite: /Users/max/.openclaw/openclaw.json (sha256 e640ced47ce8beb5b618ec100c4d78b50080c2aa6dd88ce10cbb9c6f45e89a5f -> a25347b5f9e63fbe699781ab63593e08a54d345d3614674dd5fcc5757c005981, backup=/Users/max/.openclaw/openclaw.json.bak)
Updated agents.defaults.models. Restart the gateway to apply.
1-{
2: "openai-codex/gpt-5.4": {
3: "agentRuntime": {
4- "id": "pi"
5- }
6- },
7: "openai-codex/gpt-5.5": {
8: "agentRuntime": {
9- "id": "pi"
10- }
11- },
12: "openai-codex/gpt-5.5-pro": {},
13- "openai/gpt-5.5": {}
14-}

🦞 OpenClaw 2026.5.12 (f066dd2) — We ship features faster than Apple ships calculator updates.

[old stream]

That part worked. The embeddedHarness unset warning is fine because it was already gone.

One more thing in your allowlist: openai-codex/gpt-5.5-pro still has no PI override. If it ever gets selected, it can trigger the same problem. Add the same override there too:

openclaw config set agents.defaults.models '{"openai-codex/gpt-5.5-pro":{"agentRuntime":{"id":"pi"}}}' --strict-json --merge
openclaw gateway restart

Then run /reset here and test.

If it still fails, paste the fresh lines from:

openclaw logs --limit 60 --plain | grep -i 'Requested agent harness\|model_fallback_decision\|embedded agent failed'