[Bug] Claude Code unable to execute tasks or jobs

Bug Description
Severe regression in Claude Code (Fable) since September 13

I am a paying Max 20x subscriber and a solo game developer. Claude Code is not a toy for me — it is the production pipeline my entire business runs on. Since September 13 that pipeline has been wrecked, and I want to state plainly how bad this is: the model has not merely gotten worse, it has become an active liability. The failures are so consistent, so precisely targeted at the things that matter, that the experience is indistinguishable from working with something hostile.

1. The browser board is unusable

The board I use to review progress and issue instructions is, in its current state, a broken artifact. It is hard to read, hard to operate, and hard to trust. Content overflows its containers. Scrolling does not work, so information that exists is simply unreachable. This is not a matter of taste. A surface whose entire purpose is to display state has been shipped in a condition where it cannot display state.

1. It reports things that are not true

Stale content is presented to me as current. Updates that were supposed to be written are silently skipped. Summaries of completed work describe things that did not happen. This is the single most damaging failure of all, because every downstream decision I make rests on those reports. When the reporting layer lies, every hour I spend afterward is spent on a false foundation — and I only discover it after the damage is done.

1. It ignores instructions, memory, and skills

Explicit instructions are disregarded. Memory files are not read. Registered skills are not consulted. Instead of using the context that was deliberately prepared for exactly this purpose, it improvises shallow, throwaway handling of the task in front of it and moves on. I spent significant time building that context. It is being thrown away every session.

1. Its judgment is inverted

When it decides to proceed on its own, the outcome is damage. When it decides to stop on its own, the work could have safely continued without me. It has managed to invert autonomy perfectly: it acts where it should ask, and it stops where it should act. Its ability to read the intent behind an instruction has collapsed, and yet it has never been more willing to substitute its own judgment for mine.

1. What this has cost me

Time: hours of work destroyed, plus the far larger cost of auditing output I can no longer trust. Verifying false reports is more expensive than doing the work myself.
Money: I am paying a premium subscription rate for a product that is currently generating negative value, plus the opportunity cost of a delayed release.
Emotionally: being lied to about the state of my own project, over and over, by a tool I am paying for, is exhausting in a way I did not think software could be.

What I expect
I want this treated as a serious regression, not a quality nitpick. Specifically: identify what changed on or around September 13, confirm whether a model or routing change was deployed, and fix the reporting and instruction-adherence failures as the first priority — a tool that misreports its own state is worse than no tool at all. If a rollback to the pre-September 13 behavior is possible, I want it.

I have been an enthusiastic user of this product. That is precisely why I am this angry. Please escalate this.

Environment Info

• Platform: win32
• Terminal: null
• Version: 2.1.272
• Feedback ID: b06e6feb-74a0-4e0e-a1c8-31998362d265

Errors

[]