Developers Are So Done With Claude's Word Sal

A GitHub repo called “i-have-adhd” hit 30,000 stars in September 2026. It’s 140 lines of markdown telling Claude to stop talking so much.

That number — 30,000 stars for a prompt that says “please get to the point” — tells you everything about how badly things have gone wrong with Claude’s writing quality. And a 363-comment Hacker News thread dissecting it reveals something deeper than a cosmetics problem: developers think Anthropic shipped a regression, know it’s a regression, and are watching the company scramble to fix it while still charging premium prices.

What Happened

The thread was simple. Developer domhudson posted a link to i-have-adhd, a “skill” — a markdown file you drop into Claude Code’s skill folder — that instructs Claude to put the answer first, skip preamble, and keep responses under 100 words when possible.

The framing was tongue-in-cheek: Claude Code is so verbose that you need a disorder-named workaround just to read its output. The thread title treated it as a practical tool. The comments treated it as a symptom of a much bigger disease.

One developer summed up the entire situation in two sentences: “The biggest ‘Claudism’ that I have a hard time getting the LLM to stop doing is its insistence on talking about what it didn’t do in addition to what it did. ‘I edited this.py and that.py but I did not edit README.md and I did not commit…’”

Another captured the daily ritual: “My most used follow-ups with Opus 5 are ‘one thing at a time’ and ‘too much text.’ Really hope they figure this thing out.”

Developer Reactions: It’s Worse Than You Think

The thread split into three distinct camps, and the proportions are telling.

Camp 1: “Opus 5 Is a Massive Regression”

The largest group didn’t just complain about verbosity — they called it a product failure.

“Opus 5 is such a massive regression, I really don’t understand how Anthropic green-lit it. It makes me wonder so much about the company. Like, did the people who work there actually have to suffer through its absolutely unintelligible word salad like the rest of us?”

The word “regression” appeared repeatedly. This isn’t a new model that happens to be chatty. Developers who used Claude successfully at earlier versions are watching the output quality actively deteriorate.

“Recent language Claude uses is very heavily loaded and hurts me when reading. This started only with Opus 5 where Anthropic thought they optimized the meaning per token and make it very hard to read.”

Multiple developers reported downgrading from Opus 5 to 4.6 or 4.8 specifically for readability. One wrote: “Downgrading from Opus 5 to Opus 4.8 was the biggest improvement with readability. I do use the skill in question and with Opus 5 the improvement felt marginal.” Another added: “Avoid Opus 5 — 4.6 and 4.8 work much better. 4.6 being by far the easiest to read and also just a good collaborator.”

A developer named code_biologist offered a theory that surfaced repeatedly: “Opus 4.7/4.8/5 have a built-in nitpicking as anti-sycophancy. The models are structurally incapable of not nitpicking. I suspect the nitpicking helps engagement metrics and it doesn’t harm RLVF outcomes, so there’s just not good signal against it.”

That comment — suggesting Anthropic may have trained the model to be verbose because it looks more thoughtful to casual evaluators — is the thread’s most controversial claim. If true, it means the verbosity isn’t a bug but a feature that accidentally hurts developers while looking good on benchmarks.

Camp 2: “Just Use an Output Style” (The Skill-vs-Setting War)

The second camp’s reaction to the skill was less about verbosity and more about why this needs to exist at all.

“Why is this a skill? It should be an output setting, so you can just set your output to this, and that’s it. You don’t have to run the skill every time. Go into settings and look up outputs, and just set it up there. As a skill, it’s worthless.”

Claude Code does have built-in output styles — including one called “Concise” that leads with the result, skips preamble, and keeps responses short. So why does a 30,000-star repo exist for something Claude already supports natively?

Because, as multiple developers confirmed, it doesn’t work reliably.

“I used it for a few months, and can say that no skill is up to the task of keeping Claude’s verbosity in check. You’ll get a lot more mileage out of an ‘output style’ with the same instructions, because output styles regularly remind Claude how to speak. Even output styles are not always up to the task.”

“Weak is even giving it too much credit imho, a yapping skill telling Claude to not yap can’t but fail.”

The technical explanation came from tstrimple: “CLAUDE.md doesn’t survive longer contexts because it’s towards the top of the context. One of the reasons the Claude Code voice survives is it’s part of the output style which gets ‘reminded’ to the context at every turn.” In other words, the skill sits in the prompt where it gets buried. Output styles sit in system instructions that get re-injected. Even that advantage isn’t enough.

The most pragmatic workaround? “I placed those instructions in setting > personal instructions, in my global CLAUDE.md, and in each project’s CLAUDE.md. I also use the concise output style in CC. If I choose a high thinking model for asking the weather you can’t but expect yapping.”

One developer admitted to a more visceral approach: “The most useful fix to me has been yelling at it ‘Speak English!! No jargon!!!’”

Camp 3: “Just Switch to Codex” (The Escape Hatch)

A third, smaller camp offered the simplest solution: leave.

“I solved this by switching from Claude to Codex. It’s crazy how much faster it is to reply and how much less verbose its replies are. Dunno why Claude is popular at all…”

Another noted: “I still use GPT 5.5 daily. It doesn’t seem to have this problem. It just gives me a concise summary and gets to work.”

GLM (Zhipu’s models) got the same criticism: “GLM, being trained to speak a very similar Claudeish, is also atrocious about this.” The term “Claudeish” — referring to the specific verbal style Claude developed — had become a category label for a whole class of annoying output.

The Real Insight: This Is an AI Safety Problem Disguised as a UX Bug

Buried in the thread were comments that reframed the verbosity debate from annoying to genuinely dangerous.

“Instead of telling me what to do, it first tells me 3 ways that might technically work, but I probably shouldn’t use for various reasons. This is all a waste of my time, tokens, and sanity.”

The word tokens is doing work here. Every unnecessary paragraph costs money. Claude Code charges per token. When a developer’s “concise” follow-up prompt is followed by 2,000 words of preamble before the actual answer, they’re paying to read a wall of text they didn’t ask for.

One developer calculated the behavioral cost: “Sycophancy makes me overestimate myself. When my stress meets external pressure, quality goes down. The little voice of ‘surely the AI got it right’ can lead to defects and code quality going down.”

And then there’s the signal that Anthropic is aware of the problem. Developer TomGarden wrote: “It seems they are taking it seriously now — Boris et al have been mentioning they are working on a fix and shipped a temporary band-aid output style to combat Opus 5’s horrendous prose. We’ll see if they can do it.”

A separate comment confirmed: “They botched it. Sure, but worse… they shipped it anyway. And that is the part that gets me. It’s a vastly worse product than it was on like 4.6. I suppose you can (and should) use Opus 4.6 — they do make it available still. Then just treat 5 as something you avoid until the reviews get better.”

What It Means for You

If you’re using Claude Code or any Claude-powered tool, these patterns are worth acting on:

Downgrade if you can. Opus 4.6 and 4.8 produce significantly more readable output than Opus 5. Several developers confirmed this is the single highest-impact change.
Layer your verbosity controls. The skill alone isn’t enough. Use the built-in “Concise” output style and your global CLAUDE.md and per-project instructions. Belt, suspenders, and duct tape.
Watch your token spend. If your Claude Code sessions are ballooning in cost, verbosity is likely a factor. The “speak English” follow-up might save you real money.
Don’t dismiss the regressions. If you tried Claude six months ago and found it usable, try again with the current models — but use 4.6/4.8, not the default 5.x. Your experience may have gotten objectively worse.
The competitive window is real. Codex and GPT 5.5 are being praised specifically for not having this problem. If Anthropic doesn’t fix it, the developer migration has already started.

This article synthesizes developer opinions from Hacker News thread 49610631 (363 comments, 535 points), posted September 8, 2026. All quotes are paraphrased or lightly edited from real comments. The thread remains active and continues to grow.

Developers Are So Done With Claude's Word Salad That They Built a Viral 30k-Star Workaround