#RateLimits
GitLab(.com)'s new ratelimits for unauthenticated users really is something. I looked at like, three pages, and then on my fourth I got a plaintext "Rate limit exceeded" message that isn't going away.

At least the login page isn't affected by the ratelimit.
November 14, 2026 at 9:58 AM
💬 @github-actions — PR #5798 : Add rate limiting to the OAuth provider
Add rate limiting to the OAuth provider by matthieusieben · Pull Request #5798 · bluesky-social/atproto
Commentaire de pull request · PR #5798 Add rate limiting to the OAuth provider · @github-actions No blocking findings. The PR looks fine. I reviewed the server side most closely, since the security risk is there. I also skimmed the UI changes. What I checked RateLimitManager and the policies Rules with no points always count as exceeded, since 0 - 1 < 0 . So sign-ups still require a captcha even when rateLimits is unset. That keeps the old behaviour when PDS_RATE_LIMITS_ENABLED is off. Captcha-tier rules are left out when hCaptcha isn't configured. The PAR policy rejects a captcha tier at construction. A failed block tier is reported before any hCaptcha verification runs. No block-tier default is keyed on identifierKey alone, so nobody can lock a victim out. The identifier-only keys sit in the captcha tier, where the worst case is that someone forces captchas on another person's account. Limiter names include durationMs , so reordering rules can't hand a 5-minute rule the counters of a daily one. PDS adapter ( packages/pds/src/rate-limits.ts ) xrpc-server's RateLimiter.consume throws RateLimitExceededError when over the limit and returns null when it consumes nothing. The adapter rebuilds an unclamped remainingPoints from consumedPoints , so the manager's < 0 check is correct. When the backend fails, it logs and lets the request through (fails open), as the docs say. PAR-only /oauth/authorize The new request_uri == null check throws InvalidRequestError , not AuthorizationError . So a redirect_uri that was never validated can't be used as an open redirect. The cookie-test branch now always loads the request from request_uri . The authorize() signature is narrowed to match. The only remaining 'request' in … check ( oauth-provider.ts:588 ) is on the PAR body, not on the query union that changed. Nothing else in the repo relied on in narrowing. oauthAuthorizationRequestQuerySchema : each variant uses z.never().optional() on the others' keys. A request_uri together with response_type now parses as the URI variant, and request_uri together with request is rejected, as the changeset says. hCaptcha Any verification or network failure becomes captcha_rejected , but an OAuthError thrown by the onHcaptchaResult hook is passed through. Client tokens with undefined values are dropped by JSON.stringify . Invite codes on sign-up : inviteCode now always reaches OAuthStore.createAccount , which verifies it when present ( oauth-store.ts:159 ). The changeset documents this behaviour change. Error headers : getResponseHeaders() replaces the three special cases for WWW-Authenticate , and adds Retry-After to the CORS exposed headers. UI ( RateLimitedForm ) With no Retry-After , setCountdown(undefined) falls back to 0, so the form isn't left blocked. The form only re-submits by itself when a submission was waiting for a token, so a passive challenge can't loop. Changesets : they cover every touched package at suitable bump levels (pre-v1 breaking changes are minor ). AGENTS.md is updated with the new oauth-provider rules. Minor notes, not blocking The root package.json verify script gains < /dev/null . It's a harmless tooling tweak, but it has nothing to do with the PR's stated purpose. A host-supplied rule with durationMs < 1000 would reach the PDS limiter as duration: 0 , which rate-limiter-flexible treats as "never expires". The defaults are all minutes or longer, so this only matters for custom policies. CI was still running when I reviewed, so I couldn't check its results. Claude review, run Ouvrir sur GitHub →
github.com
November 18, 2026 at 2:03 PM
afaik theyre just records on ur repo so ur only limited by bsky's firehose/relay per pds ratelimits i think
November 9, 2026 at 4:10 PM
Oh the mechanics.
November 2, 2026 at 11:33 PM
if this is at all a serious report and not just more trolling horseshit, yeah now thanks to A/I anyone including a single teenager can just pull stuff like this across the entire internet.
i need a drink
November 2, 2026 at 11:17 PM
It's down.

but also, technically it was fun??? Read the fucking room dude...
November 2, 2026 at 11:30 PM
WELP. i was wrong about this one, they were BOTH well-meaning "fans" and this one was 17. and where did his swarm of AI agents get all of these fics? that's right, publicly-shared google drives!!!!
November 3, 2026 at 12:02 AM
It's down.

but also, technically it was fun??? Read the fucking room dude...
#ao3
November 2, 2026 at 11:31 PM
That was fast.

He's full of shit, but whatever. Site's dead.
November 2, 2026 at 11:18 PM
i'm keeping this up bc it's a fascinating look into how these quick-build vibecoded AI slop sites function, with the addition of AI agent swarms
November 3, 2026 at 12:15 AM
@tangled.org are org creations broken currently, or am I the problem 🥺
October 31, 2026 at 5:00 PM
essa criança aqui sendo irmã do filho da policial de esquerda
November 3, 2026 at 12:20 AM
Something I really want to come to atproto is the ability to do something like `deleteCollection` to bulk delete an entire collection of data in one go, instead of having to constantly iterate through records one by one and delete them whilst abiding by ratelimits of the PDS and relays
October 29, 2026 at 6:12 PM
mlem r9k feed is back up and iqdb feeds should be up soon (labels maybe not so... label ratelimits are too tight.. if anyone wants to build a client with labels built in, feel free to hit my queryLabels)
October 29, 2026 at 10:10 PM
On the infrastructure side, commenters noted how futile robots.txt and IP blocks are against corporate scrapers. Blocking AWS or Azure ranges doesn't stop anyone; teams just route through residential proxy pools to bypass ratelimits anyway. #devops 3/4
October 27, 2026 at 9:00 PM
GitLab kept rate-limiting me, so I blamed my twenty agent sessions and rebuilt how they authenticate. They'd been idle all night. The trace found the culprit: me, opening four merge requests at once.

#RateLimits
I blamed the agents for my rate limit
How a trace of GitLab's RateLimit headers cleared twenty idle agent sessions and pinned the 429s on the human clicking merge request links.
phpboyscout.uk
October 29, 2026 at 10:12 AM
Cómo resumir texto en salud con Chat Completions API Node.js

Aprendé a resumir texto con la Chat Completions API Node.js para armar bases de conocimiento privadas en salud. Descubrí el ejemplo completo acá.

#chatcompletionsapi #nodejs #openai #healthtech #ratelimits
Cómo resumir texto en salud con Chat Completions API Node.js
Código real en Node.js/TypeScript para resumir documentos clínicos con la Chat Completions API: temperatura, reintentos y qué no perder en el resumen.
blog.donweb.com
October 25, 2026 at 2:31 AM
October 11, 2026 at 10:59 PM
Added time-based ratelimits, let me know if you guys hit em by accident.
October 10, 2026 at 10:26 PM
i was one thing i can post until iceshrimp itself ratelimits me
October 6, 2026 at 11:43 AM
Sharkey ratelimits me on WebKit is incredibly efficient and I had to watch that little blood sucking stick for 3 extra minutes
September 28, 2026 at 10:20 PM
The 50 percent weekly boost for Claude Code ends September 13. From today, limits settle 17 percent below current levels, a cut Anthropic itself published.

#AiCoding #Anthropic #ClaudeCode #Developers #OpenAI #RateLimits
Claude Code Limits Drop Today as Anthropic Reset Takes Effect
The 50 percent weekly boost for Claude Code ends September 13. From today, limits settle 17 percent below current levels, a cut Anthropic itself published.
pulseofnations.lol
September 14, 2026 at 7:48 AM
ChatGPT Plus — Codex 5-hour rate-limit bucket missing on my account
My ChatGPT Plus account is missing the Codex 5-hour usage window. The Codex app and account/rateLimits/read both return only the weekly bucket: primary: { usedPercent: 7, windowDurationMins: 10080 } secondary: null ← the 300-minute bucket does not exist planType: plus credits: balance 0, hasCredits false rateLimitResetCredits: availableCount 0 To be clear, this is not a general entitlement failure on the account: GPT-6 Astra became available in my model picker on the same day. The only missing item is the 300-minute rate-limit bucket. Requests: 1. Check the rate-limit configuration attached to my account and confirm whether a 5-hour window should be active. 2. If it should be: fix that configuration. 3. Restore usage that was absorbed by the weekly allowance while the 5-hour window was absent, and/or grant reset credits. Evidence: local rate-limit snapshots show both buckets present on 2026-09-07 21:57:09 (300 min + 10080 min), and only the weekly bucket from 2026-09-12 18:12:11 onward. The client was updated four hours later, so this is a server-side change, unrelated to the client version. Account ID: 201cf045-3c12-4de1-9772-f448f4f48b87 Codex desktop 26.903.8094.0 / CLI 0.153.4 / Windows 11 26200 Feedback ID: [FEEDBACK_ID]
community.openai.com
September 12, 2026 at 2:12 PM
unit tests would be really good to have but i still have yet to come across any type of system that would work reasonably well given that it is a bot for a centralized chat platform with heavy ratelimits
September 11, 2026 at 3:53 AM