#CreateAccount
I'm all for it. I wasn't a fan of using the PDSs admin password for it lol.

Any chance to get the need for a plcOp on createAccount relaxed if the pds is in entryway mode 👀
November 12, 2026 at 7:58 PM
November 12, 2026 at 4:32 AM
Merged PR into main at bluesky-social/atproto: pds: accept the entryway service JWT on createAccount (#5847)
November 9, 2026 at 2:14 AM
Imo, createAccount shouldn't even exist, which would also allow describeServer to be cleaned up.
November 6, 2026 at 10:43 PM
pds: accept the entryway service JWT on createAccount (#5847)
pds: accept the entryway service JWT on createAccount (#5847) · bluesky-social/atproto@0849c24
pds: accept the entryway service JWT on createAccount (#5847) Behind an entryway, createAccount is called by the entryway alone, with no authentication. Its user service auth verifier cannot take the entryway's service JWT: it looks the key up in the issuer's DID document, and the entryway's did:web does not resolve. In entryway mode the route now verifies a bearer token as an entryway service JWT, as reserveSigningKey does, and lets any other request through unauthenticated. A valid entryway JWT for createAccount already bypasses rate limits, including the route's own, through the attribution middleware. Unlike the other entryway-called routes, PDS_ENTRYWAY_REQUIRE_SERVICE_AUTH does not make this route refuse unauthenticated calls yet: PDSes run with it on, and the entryway can only start signing createAccount once they accept a signed call. Enforcement follows once it does. Only user service auth names a requester, since the entryway's credentials name the entryway. The entryway-mode checks of the PLC operation, its rotation key and the reserved signing key are unchanged, as is the route without entryway configuration. Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
github.com
November 9, 2026 at 2:25 AM
mood
November 5, 2026 at 7:47 PM
🟢 PR #5891 — pds, xrpc-server: scratch Redis failures fail open
pds, xrpc-server: scratch Redis failures fail open by amckinley · Pull Request #5891 · bluesky-social/atproto
Pull request (OPEN) · PR #5891 pds, xrpc-server: scratch Redis failures fail open · @amckinley A PDS's scratch Redis holds rate-limit counters, DPoP replay records and a scope cache. Losing it should cost that state, not requests. Three paths didn't fail open: Route rate limits (xrpc-server). A route's own limit was created without onError , so a store error failed the request with a 500: createAccount , createSession , sync.getRepo , uploadBlob , updateHandle and the rest, on any PDS whose Redis goes down. Global and shared limits already logged the error and let the request through ( failClosed defaults to false). Route limits now do the same, unless they set the new failClosed option. Key prefixes don't change, so existing counters carry over. The DPoP replay check (PDS). ReplayStoreRedis rejects when Redis fails, so every OAuth request failed. The PDS now passes a FailOpenReplayStore , which uses the same keys and also keeps each process's own history: a nonce either has seen is a replay. While Redis is down, replays to the same process are still refused, including proofs first seen before the outage or after recovery. Replays to another process aren't caught until Redis is back. No timeouts (PDS). The scratch ioredis client queued commands while disconnected (about 10 s of retries per command) and waited forever on a Redis that stopped answering. It now fails a command after 500 ms, or at once when disconnected. Shutdown drops a connection that won't answer QUIT instead of failing. The scope reference cache already failed open ( swallowStoreErrors ). Tests: xrpc-server: a rate-limit store that always errors lets requests through a route limit, and fails a route that sets failClosed . The first test fails on the old code. PDS: unit tests for the replay store, including replays across Redis failing and recovering, and for the client's timeouts against a hung Redis and an unreachable one. PDS end to end: a PDS whose Redis completes the handshake and then never answers still creates accounts, writes and serves reads. Before the change, createAccount returned 500. The xrpc-server suite (181 tests) and the PDS rate-limits, auth, entryway service auth and account suites (169) pass. 🤖 Generated with Claude Code Ouvrir sur GitHub →
github.com
November 13, 2026 at 2:38 AM
a generic solution/interface. eg. the inviteCode field couldve been used for a lot of the verificationCode stuff instead i think. as a generic "extra thing you need to pass to be allowed in" or the verificationCode couldve been a ref impl specific extension to createAccount instead
November 6, 2026 at 10:32 PM
to be fair im being a bit hyperbolic about it. but eg. describeServer has phoneVerificationRequired and similarly createAccount has verificationCode and verificationPhone (and email which i dont want there as part of "i dont think the protocol should assume or dictate email"). i do get some of these
November 6, 2026 at 10:32 PM
tbh createAccount getting ditched in favour of prompt=create + a theoretical group.opensocial.createGroup or something (and whatever implementation specific thing people might want from there) would probably cover everything i care about
November 6, 2026 at 10:57 PM
As part of my occasional posts on #poorux today I'm trying to register with the #openuniversity and it's not going well.

This is the URL of the form
https://www.open.ac.uk/account/createaccount/

First issue: they require your title, from a pull down list, and of course it's gendered unless you […]
Original post on mastodon.scot
mastodon.scot
November 7, 2026 at 1:07 PM
I think so too.

@angrydutchman.online, I know you're doing create via OAuth, but may also look at doing it via password session on createAccount. It has a param for passing a recovery key (rotation did:key) on submit and can do that client side

endpoints-standalone.onrender.com#bluesky-app/...
Bluesky HTTP API Reference
endpoints-standalone.onrender.com
October 23, 2026 at 12:42 PM
Yeah, but tbh, the createAccount XRPC call should be deprecated.
October 22, 2026 at 5:04 PM
I've got it to the point now where creating an account via the `prompt=create` path will pretty much do this, the only place you can't force it (well...) is via a `createAccount` call - unless you make the `recoveryKey` field mandatory which, well, I'm not sure how many things that'll break 🤣
October 22, 2026 at 5:03 PM
自分の見落としかもしれないけど、公式のPDS実装はcreateAccountを防ぐ方法が無くてスパムアカウント作り放題になるので、mochottとか用のAtmosphereアカウント新規作成の基盤のためにPDSから作るしかなかった
October 6, 2026 at 12:11 PM
idk, just get mediawiki and set these options mayb
August 31, 2026 at 2:37 AM
This is only a problem for multi-user PDSs where the createAccount endpoint is not gated by invite codes. It looks like your software (micropod?) is single-user, so it's not impacted. Same with mine. My PDSs don't even *implement* createAccount.
March 25, 2026 at 6:51 PM
Then click "Generate key". It'll give you a big thing of JSON that you need to make accessible at "https://coolguy.example.com/.well-known/did.json". Then click "Sign createAccount". You'll get a long cURL command you can copy and paste.
November 26, 2024 at 4:22 AM
You're giving away the secrets!

I didn't actually find this tool until a few hours into figuring it out for myself based on another thread somewhere; had quickly thrown together a script to do the createAccount calls and dump the new DID doc from the API
November 25, 2024 at 7:52 AM
stays in your browser, only used to sign the createAccount request
November 26, 2024 at 5:42 PM