#my OC on codex sudenly went lazy and dumb

1 messages · Page 1 of 1 (latest)

wide tangle
#

Anybody have same experience?
i have codex on open ai plus oauth.
sudenly it just promising "i will do it", but don't deliver. then it stars to say "sorry, i will fix it" but doesn't.

it is just me or others having same experience?

it started yesterday.

next vigilBOT
#

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

young jewel
#

i have same problem sice yesterday night.

#

he always say he will do. but doing nothing in the and and alwys say sorry i will do. but nothing changed. he makes me crazy...try a lot of things but nothing helped

autumn narwhal
#

having same issues

young jewel
#

someone find a trick or solution to solve this crap?

#

i try already to downgrade version, but not help. he start getting lazy and dumb yesterday evenig same time arround i do the new update. because this i try to downgrade version but nothing change. i doublecheck every single skill for updates and also node, restart again and again gateway and skip all cronjobs. but still nothing

young jewel
#

@worldly ocean you maybe know what to do?

young jewel
#

Bug type

Regression (worked before, now fails)

Summary

After updating to OpenClaw 2026.3.23-2, openai-codex/gpt-5.3-codex stopped executing any tools. The agent acknowledges the request and says it will execute, but then does nothing — no tool call is ever made.
When switching to the Claude Opus model over API call, it works.

Environment:

OpenClaw: 2026.3.23-2
OS: Ubuntu 22.04 (VPS)
Node: 22.22.0
Model: openai-codex/gpt-5.3-codex via OAuth (ChatGPT subscription)
Channel: Telegram
tools.profile: full

Steps to reproduce

Set primary model to openai-codex/gpt-5.3-codex with tools.profile: full
Ask agent to execute any tool — exec command, MCP Gmail, MCP Google Calendar, web search
Agent responds with text only, never calls any tool

Expected behavior

Agent executes the tool and returns results

Actual behavior

Actual: Agent says it will execute but produces only a text response with no tool calls. Example responses:

"I'll check your calendar now" → returns nothing
"Running gog auth now" → does nothing
"I'll check the cron jobs" → returns fabricated text instead of real data

spare crater
#

Same issues here, any new ideas on what it could be?