[Bug] Agent validation checks pass incorrectly on broken deployments

**Bug Description**

❯ What's the point of having the model run a check for deployment confirmed on a gated live version if its just going to anyway confirm everything as working when they are actually broken. Wastes shit load of time. Also, what's the point of approving a plan when it is not not going to follow it?? I don't care if the agents did not work as specified, its the orchestrators job to check and get things corrected/rectified, I don't pay $200 a month to Anthropic for this shitty service, and off late the instances have only been increasing across projects. At this rate 1 more such incidence and I'll surely cancel/not renew the subscription. Codex is far far better but I've so much of configuration built in Claude that its a time consuming effort to switch to Codex, but I guess I have to do it now anyway as this is not working and every project is starting to get stuck in advanced stages and the models are failing to comprehend the depth of the projects and start hallucinating/taking shortcuts. And all this is happening with models like Fable 5/5.1 and Opus 4.8. Sonnet 4.5 was way better due to the time it was in the market and the sheer amount of coding inputs it probably went through working on tasks but you ruined that as well with Sonnet 5. Not even sure why I am using Claude anymore.

**Environment Info**
- Platform: win32
- Terminal: windows-terminal
- Version: 2.1.268
- Feedback ID: 52281101-0ddb-468a-bce7-7ea682d1928b

**Errors**
```json
[]
```