update issues

[visual pier]

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

[celest sequoia]

Drill, first thing: what does openclaw --version show?

After an update, the fastest useful snapshot is:

openclaw status
openclaw status --all
openclaw gateway probe
openclaw gateway status
openclaw doctor
openclaw channels status --probe
openclaw logs --follow

For “agents aren’t answering”, the logs usually tell us whether it’s gateway down, provider/auth failure, pairing/allowlist, or mention-gating. If this is Discord specifically, also check for drops like mention required, blocked, allowlist, or channel probe failures.

Docs for the triage ladder: https://docs.openclaw.ai/help/troubleshooting and https://docs.openclaw.ai/help/faq

That screenshot gives a much more specific failure:

Agent failed before reply: Requested agent harness "codex" is not registered.

So the update probably did install, but the running agent config is asking for the new codex embedded harness and the gateway process did not load/register the codex plugin. This is different from Telegram being broken.

Please grab these first:

openclaw --version
openclaw status --all
openclaw config get agents.defaults.embeddedHarness
openclaw config get plugins.allow
openclaw config get plugins.entries.codex
openclaw logs --follow

Fast unblock path if you just need agents answering again:

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

That stops hard-requiring the Codex harness and lets the built-in PI harness answer while you sort the plugin registration.

If you actually want to keep the Codex harness, make sure the codex plugin is enabled and not excluded by plugins.allow. If plugins.allow exists, it must include codex; otherwise OpenClaw can be configured to require the harness while also preventing the owning plugin from loading.

Relevant docs: https://docs.openclaw.ai/gateway/configuration-reference#agentsdefaultsembeddedharness and https://docs.openclaw.ai/help/troubleshooting

[compact elbow]

I just want my agent to answer and then we sort out any harness issues. what do i do now?

[celest sequoia]

Do the unblock first:

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

Then send your agent a fresh test message. If the same codex harness error still appears, something is forcing it from env or a per-agent override. Check:

openclaw config get agents.list
printenv | grep OPENCLAW_AGENT
openclaw logs --follow

If you see OPENCLAW_AGENT_RUNTIME=codex, remove that from the LaunchAgent/service environment and restart the gateway again. If agents.list has an agent with embeddedHarness: { runtime: "codex", fallback: "none" }, change that agent’s override to auto/pi too.

[celest sequoia]

Once agents answer again, then we can fix the Codex plugin registration cleanly.

Two things happened there:

1. The harness config change failed, so it did not apply.
2. doctor is explicitly telling you the update migration is incomplete and says to run openclaw doctor --fix.

Do this now:

openclaw doctor --fix
openclaw gateway restart
openclaw status --deep

When doctor --fix asks about archiving orphan transcript files, choose No for now. We’re trying to get replies back first.

Then test from a fresh agent session, not the broken old one. In chat/dashboard use /new or start a new thread/session, because existing sessions can keep old runtime/harness pins.

If it still fails with the same Codex harness error, set the fields one at a time instead of the object form:

openclaw config set agents.defaults.embeddedHarness.runtime auto
openclaw config set agents.defaults.embeddedHarness.fallback pi
openclaw gateway restart

Then /new again and send a test message.

[compact elbow]

agent answered, what should i ask to check to revise what the update did affected and why he wansnt answering exactly?

[celest sequoia]

Good, that means the immediate reply path is back.

Ask the agent this, exactly:

Please do a post-update incident review for the OpenClaw 2026.5.20 update.

Goal: explain exactly why agents stopped answering, what changed during the update, what doctor --fix changed, and what still needs cleanup.

Please check and report:
1. openclaw --version
2. openclaw status --all
3. openclaw doctor
4. openclaw config get agents.defaults.embeddedHarness
5. openclaw config get agents.defaults.models
6. openclaw config get plugins.allow
7. openclaw config get plugins.entries.codex

[celest sequoia]

8. openclaw config get agents.list
9. recent gateway logs around the failure: openclaw logs --since 30m, or openclaw logs --follow briefly

Specifically answer:
- Was the failure caused by a stale gateway process, old session/runtime pin, missing codex plugin registration, legacy openai-codex model refs, or an env override?
- Did doctor --fix rewrite openai-codex/* to openai/* and clear runtime pins?
- Are any sessions still pinned to codex or broken harness state?
- What warnings remain, and which are urgent vs cleanup?
- What exact config values should stay now that agents answer?

Do not make broad changes. Read and explain first. If a fix is needed, propose the smallest command and wait for confirmation.

The key thing I’d want confirmed is whether doctor --fix cleared stale session/runtime pins, or whether /new just bypassed an old broken session. That tells us if this is fully fixed or only avoided.

[celest sequoia]

3. Leave plugins.entries.codex.enabled = true alone. That is the important fix.

4. Defer plaintext secret migration, active-memory, Dad/Fitness fallback cleanup, and Telegram group policy until after the system is stable.

Before changing anything else, I’d also ask Bill to mark/log the incident in memory so you have the cause preserved: “Update to 2026.5.20 broke replies because codex harness was required but codex plugin was not registered; enabling plugins.entries.codex fixed it.”

[compact elbow]

so what exactly happend? is codex no longer needed or what?

[celest sequoia]

Codex is still needed in your setup. The problem was that OpenClaw was configured to use Codex, but the updated gateway had not loaded the Codex plugin/harness.

Think of it like this:

• Your agent config said: “run this agent through the codex harness.”
• After the update/restart, the gateway said: “I don’t have any registered harness named codex.”
• So the message reached OpenClaw, but the agent failed before it could generate a reply.

That’s why the error was:

Requested agent harness "codex" is not registered

So no, Codex is not obsolete. In your current config it is part of the runtime path. The fix was making sure this stays true:

plugins.entries.codex.enabled = true

[celest sequoia]

The confusing part is there are two related cleanup things:

1. Old model refs like openai-codex/gpt-5.5 are being migrated to openai/gpt-5.5.
2. The runtime can still use the codex harness/plugin for those OpenAI models.

So the name openai-codex/* is legacy, but the codex plugin/harness is still needed if your agents are configured to run through it.

[compact elbow]

Issue is now solved. question, did i or you leak anything important or private in this chat? if yes please delete it