Skip to content
topics.

Anthropic publishes Claude Code post‑mortem and issues fixes

devApr 23, 20262889

Anthropic published a post-mortem and shipped fixes in v2.1.116+, and it reset usage limits for all subscribers. The company identified three root causes: the default reasoning effort had been lowered from High to Medium; a caching bug that erased Claude's memory mid-session; and a system prompt change that increased verbosity and reduced coding output quality. Anthropic traced the problems to Claude Code and the Agent SDK harness, which also affected Cowork because Cowork runs on the SDK, while the Claude API and the underlying models were not affected. With the fixes deployed in v2.1.116+, Anthropic says code-generation behavior should be restored in Claude Code and SDK-based products.

Djinn & Tonic 🇺🇦 🇨🇦 🇲🇽 🇬🇱
@djinnandtonic.bsky.social

I built a duck feeding robot out of a solar panel, a marine battery, an ATV winch, and a microcontroller that runs on Claude code, works good Only issues I ever had have been mechanical, which isn't really Claude's issue, that was me failing to account for winch-related stuff

333h ago
Dare Obasanjo
@carnage4life.bsky.social

Anthropic has tracked down and fixed why Claude seems to have gotten dumber in the past month. 1. Reduced default reasoning effort (from High to Medium). 2. A caching bug that accidentally deleted Claude’s memory mid-session. 3. A system prompt change for verbosity that hurt coding ability.

293h ago
Techmeme
@techmeme.com

Anthropic says it has fixed three causes of recent Claude Code quality issues: reduced default reasoning, a caching bug, and a system prompt to reduce verbosity (Anthropic) Main Link | Techmeme Permalink

63h ago
Phillip Carter
@phillipcarter.dev

first issue seems to comport with people on here feeling 4.7 vibes are off, I wonder if we'll see another update trying to further balance the tradeoff between final code output quality and interactivity www.anthropic.com/engineering/...

23h ago
ellie lockhart (she/they)
@eleanor.lockhart.contact

How is Letta Code, as an alternative to Claude Code, these days? Do we have an actually solid implementation of something like OpenClaw with it yet?

43h ago
ClaudeDevs [UNOFFICIAL]
@claudedevs-mirr.selfhosted.social

Over the past month, some of you reported Claude Code's quality had slipped. We investigated, and published a post-mortem on the three issues we found. All are fixed in v2.1.116+ and we’ve reset usage limits for all subscribers.

14h ago
Gilbert Sanchez (💻 HeyItsGilbert)
@gilbertsanchez.com

They finally published the claude code marketplace schema to SchemaStore! github.com/anthropics/c... You guys know how I feel about json schemas.

Gilbert seeing a proper json schema. Using the "Feels Good Man" meme template
33h ago
cloud or poo?
@mullets.bsky.social

The deepening self-suck of AI investment. "Some #Google DeepMind employees have been permitted to use @anthropic.com Claude Code in recent months, which has created some tensions between employees, BI previously reported. Google certainly isn't the only tech giant turning more to AI for coding." 🧵

24h ago
Robby Russell
@robbyonrails.com

You gave your team AI coding tools. Output went up. So did variance. Same tool, same codebase... completely different approaches. There's still time to get your team aligned on this. robbyonrails.com/articles/202...

23h ago
Erorus
@eror.us

I also tripped over a post about Qwen3.6-27B, the latest hot shit local coding model or whatever. I have an M4 with 24GB, so why not follow some random gist to set it up with Claude Code. For science. After some OOM issues and tweaks, it runs. I've unleashed it to refactor my 250-line project.

14h ago
3 sources
theregister.comClaude Code routines promise mildly clever cron jobsanthropic.comAn update on recent Claude Code quality reportst.fxtwitter.comClaudeDevs (@ClaudeDevs)

Related

Anthropic publishes Claude Code post‑mortem and issues fixesdevUbuntu 26.04 LTS, 'Resolute Raccoon' releaseddevBitwarden CLI 2026.4.0 compromised in Checkmarx supply chain attackdev