THIS GUY LOST $200 IN ONE DAY BECAUSE THE STRING "HERMES.md" WAS IN HIS GIT COMMITS HERMES.md is a real convention used in AI agent projects. it's a system prompt specification file. not some obscure edge case he's on claude max 20x at $200 a month. yesterday claude code hit him with "you're out of extra usage" out of nowhere his dashboard showed 13% weekly usage. 0% current session. 86% of his plan was sitting there untouched but $200.98 in extra usage already burned through what should have been covered by his subscription he tried logout & login, different models, fresh installs and nothing worked anthropic support sent the ai bot (four rounds of the same scripted response). eventually they just gave up on him so he started binary searching repos and commits manually on his own time until he found the trigger the string "HERMES.md" in a recent git commit message uppercase, with the .md extension, anywhere in your commit history that's it claude code includes recent commits in its system prompt and something server side flags HERMES.md and quietly routes you off your max plan onto API rate billing > AGENTS.md? fine > README.md? fine > HERMES without .md? fine > lowercase hermes.md? fine > uppercase HERMES.md? you're getting charged API rates he reported it. anthropic support acknowledged the bug three times, called it an "authentication routing issue", thanked him for finding it then refused to refund the $200 so the man pays $200 a month for max, lost another $200 to a billing bug they confirmed, did anthropic's QA work for free on his weekend, and got a "thank you for your patience" in return check your commit history before claude code quietly drains your account too
180
336
4,577
1,459,529
Replying to @om_patel5
ugh sorry this was a bug with the 3rd party harness detection and how we pull git status into the system prompt we're reaching out to affected users and giving them a refund + another month of credits (in this case another $200)

Apr 26, 2026 · 8:12 PM UTC

162
23
1,278
123,544
Replying to @trq212 @om_patel5
1. It is starting to feel like a pattern w/ anthropic, esp given past actions 2. Regardless you're doing really well in owning up when there are mistakes 3. At some point the goodwill from the latter will stop offsetting the hostility from the former At that point people will assume default-misanthropic, which is something to be very careful to avoid. Good luck.
341
Replying to @trq212 @om_patel5
Recently, I've been seeing API charges even though I'm on the Max subscription plan. What does this mean?
1
3
1,123
Replying to @trq212 @om_patel5
There’s a certain class of bugs that suggests the thing you’re trying to do is a bad idea. Worth reflecting on that here imo
15
25
1,598
38,891
Replying to @trq212 @om_patel5

ALT You Dont Say Frowning GIF

3
1,657
Replying to @trq212 @om_patel5
Why wouldn’t you guys start having a better support system instead of this automated system or public X accounts?
4
3
54
3,803
Replying to @trq212 @om_patel5
Is end_conversation a bug? It was used on me last night by Sonnet (Claude Desktop) when I said I would look at code it drafted in the morning. Locked out
1
1
1,958
Replying to @trq212 @om_patel5
@trq212 how come yall handing out credits. When I legit DM'ed you multiple times for over 18 fraudulent charges from last month and 4 from this month on ONE Claude Max Plan totaling over $ 4K in charges and get crickets? Dealing with chargebacks is bs. FinAgent is really shit.
1
1,859
Replying to @trq212 @om_patel5
so the feature is intentional, this is just a false positive
7
3
387
8,031
Replying to @trq212 @om_patel5
At this point, I just hate claude and anthropic. Incredibly dishonest and non-transparent. This is not the first "bug" that people have reported either.
1
992
Replying to @trq212 @om_patel5
You can always hire a security engineer to make an actual system that can’t be abused by third party harnesses
1
1
2,253
Replying to @trq212 @om_patel5
I am one of those who cancelled their max Anthropic sub and am currently trying the codex pro tier. News like this confirms my decision. Why be antagonistic towards people using oauth on anything they choose to use it for? The OpenClaw ban is what really got me, even tho I am mostly using the CLI on any provider. It just feels aggressive toward users and I don’t like to support
7
83
5,515
Replying to @trq212 @om_patel5
so im getting a refund?
2
15
1,576
Replying to @trq212 @om_patel5
Why does this need to go viral for someone to get a refund for a legit bug 😅
200
3,437
Replying to @trq212 @om_patel5
Y’all are absolutely doing too much if you think this was remotely ok. Don’t mess with people’s money please. I’m sure you’re doing fine without trying to extra charge people.
841
Replying to @trq212 @om_patel5
your team must be incredibly detached from reality if you think this is okay in the first place. you sell a closed source typescript terminal agent and you don't even have the best coding models anymore. wake up and reconsider whatever company culture facilitates this nonsense.
7
1,255
Replying to @trq212 @om_patel5
Hey, this happened to me and I wasn't even using automated Hermes... I was just using Claude Code to help build a bridge and created a PRD with the name HERMES in it and was locked out of using my Max account when I turned off the unauthorized extra usage. Support was no help.
1
4
921
Replying to @trq212 @om_patel5
Compensation is required where it's due, atleast lately you and Claude team aware of it's bugs and compensating, that's great thing to hear imo.
1
1,912
Replying to @trq212 @om_patel5
That is the result when you try to make a feature of your product which nobody wants and which you make your product worse because you lost the focus on your costumers
1
2,188
Replying to @trq212 @om_patel5
Your website told me "billing failed" 4 times in a row when I tried to buy a gift sub. But it was secretly taking payment from my existing credits. I never got the gift sub, not that I was even trying to purchase 4 of them. It's been weeks. No sign of a "human agent".
4
987
Replying to @om_patel5
There is zero reason for the harness to be reading ANYTHING about my project here @trq212 . Like, you're doing active FORWARD PROCESSING. That's adversarial behavior at best. It shouldnt ever matter if I'm working on a harness or another project that is named Hermes or Openclaw. All the harness should be doing is executing my request, not deciding on what I'm doing with it, so long as I'm not powering another product with it.
1
1
12
981
Replying to @trq212 @om_patel5
unbelievable. “3rd party harness detection” - you mean basic string matching. give me a break. also, stop fucking with the system prompts, we want a stable tool for professional work, every model knows how to check git history if it needs to.
1
14
601
Replying to @trq212 @om_patel5
It happened to me when all I use is Claude code CLI
744
Replying to @trq212 @om_patel5
Can i have access to mythos for the trouble, we need to close any hidden security gaps in Hermes sire 🥹🥹
5
215
6,630
Replying to @trq212 @om_patel5
Maybe right now isn’t the best time to light even more community trust on fire?
380
Replying to @trq212 @om_patel5
"sorry, we have a system to detect if you're using AI on our AI service."
17
1,697
Replying to @trq212 @om_patel5
The fact that you even have the concept of punishing 3rd party harnesses is the issue.
2
25
587
Replying to @trq212 @om_patel5
This is single handily the largest non utility bill I pay. It’s crazy how often this kinda stuff happens
1
673
Replying to @trq212 @om_patel5
Anthropic: "Hey Claude, Here at Anthropic we are running out of funds quick! Any solutions?" Claude: "Hold my beer..... i just enabled Extra usage on a random set of users"
1
275
Replying to @trq212 @om_patel5
Someone should write something in the GH bug report, people are reading this and getting all sorts of ideas: github.com/anthropics/claude…
1
645
Replying to @trq212 @om_patel5
Asking from a privacy point of view. How exactly does Anthropic identify the impacted users?
1
701
Replying to @trq212 @om_patel5
I’m glad you are fixing this and trying to make it right. I get that bugs happen, but customer service not defaulting to an immediate refund and service credit is really bad. If Theo hadn’t put a spotlight on this issue, what would have happened? GG Theo!
1
7
608
Replying to @trq212 @om_patel5
but, but wasnt software engineering over like your CEO said? why do you have so many bugs? claude cant fix them all?
827
Replying to @trq212 @om_patel5
You guys need to do some A/B testing. Shiping YOLO to peod causing bunch of bugs and AI slops now. Like recent CC regressions took months to figure out
2
195
Replying to @trq212 @om_patel5
This barely offers any reassurance, much better to say for bugs like this we will always refund & we are updating our policies.
396
Replying to @trq212 @om_patel5
So spying on us so we use it the way You want us to use it. Cool. Boy am I glad for local llms.
1
79
Replying to @trq212 @om_patel5
cost me 130.00 so wtf. no email or refund here
1
21
1,342