#My agent with GPT5.4 keeps doing the "if you want I can do this" without ever actually do anything

1 messages · Page 1 of 1 (latest)

shy dock
#

My agent with GPT5.4 keeps doing the "if you want I can do this" without ever actually do anything non matter how many times i say ok proceed implement etc

wheat pivotBOT
ionic valleyBOT
#

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

wheat pivotBOT
south solsticeBOT
#

Yeah, that’s a real failure mode GPT-5.4 had, “good plan, no action.”

First, paste:

openclaw --version
openclaw status --all

Then check these 3 things:

  1. Model route matters
    The stricter “act, don’t just plan” contract is for direct openai/* and openai-codex/* GPT-5 runs. If you’re on a proxy route like OpenRouter, you may not get that behavior.

  2. Make sure you didn’t opt out
    If your config has this, it keeps the old looser behavior:

# 
   agents: {
     defaults: {
       embeddedPi: { executionContract: "default" },
     },
   }

If needed, set:

agents: {
  defaults: {
    embeddedPi: { executionContract: "strict-agentic" },
  },
}
  1. It may be waiting on approval