alice.mosphere.at / perlsky
perlsky is a Perl 5 implementation of an AT Protocol Personal Data Server.
13
fork

Configure Feed

Select the types of activity you want to include in your feed.

Perl 99.0%
JavaScript 1.0%
218 1 2

Clone this repository

https://tangled.org/alice.mosphere.at/perlsky https://tangled.org/did:plc:by3jhwdqgbtrcc7q4tkkv3cf/perlsky
git@tangled.org:alice.mosphere.at/perlsky git@tangled.org:did:plc:by3jhwdqgbtrcc7q4tkkv3cf/perlsky

For self-hosted knots, clone URLs may differ based on your setup.

Download tar.gz
README.md

perlsky#

perlsky is a Perl 5 implementation of an AT Protocol Personal Data Server.

Current direction:

  • Official com.atproto.* lexicons are vendored into share/lexicons.
  • The external XRPC surface is loaded from those lexicons at runtime.
  • Account, repo, blob, sync, CAR, DAG-CBOR, CID, and MST support are being implemented in native Perl.
  • The app is designed to run self-contained with SQLite and filesystem blob storage.

The immediate goal is a PDS that is pleasant to hack on and interoperable enough to be exercised with real AT Protocol clients and repo sync tooling.

Developer-oriented notes about the current facade/module split live in docs/CODE_STRUCTURE.md.

Reference differential validation:

  • Run script/differential-validate to compare perlsky against the official published @atproto/pds on a focused set of account, repo, moderation, sync, firehose, and importRepo snapshot-restore behaviors.
  • The differential harness also configures a local relay/crawler mock for both servers and verifies that both emit com.atproto.sync.requestCrawl notices with the expected hostname after repo activity, based on the upstream crawler wiring in packages/pds/src/crawlers.ts, context.ts, and sequencer.ts.
  • Run PERLSKY_DIFF_ACCOUNT_DID_METHOD=did:plc script/differential-validate to exercise the same harness in PLC-account mode, including recommended DID credentials, PLC signature requests, PLC handle updates, token-gated PLC signing behavior, and moderation checks after PLC handle changes.
  • The helper installs the reference runtime into .tools/reference-runtime with Node 20 via fnm.
  • Run PERLSKY_RUN_REFERENCE_DIFF=1 prove -lv t/reference-differential.t to exercise the same harness from the test suite.
  • Run PERLSKY_RUN_REFERENCE_DIFF=1 prove -lv t/reference-differential-plc.t to run the PLC-specific reference comparison from the test suite.

Metrics and observability:

  • perlsky now exposes Prometheus-compatible metrics at /metrics.
  • Set metrics_token to require Authorization: Bearer <token> for scrapes.
  • The main runtime signals cover XRPC request counts/latency, websocket subscriptions and emitted frames, crawler notifications, blob ingress/egress bytes, and key store operation timings.
  • Detailed operator documentation lives in docs/METRICS.md.

Deployment and first-account setup:

  • Generic single-node deployment instructions live in docs/DEPLOYMENT.md.
  • The deployment guide includes a reverse-proxy layout, a sample systemd unit, validation commands, and a createAccount example for bootstrapping the first user.
  • perlsky now includes a built-in ATProto OAuth provider surface, so modern third-party clients that use the Bluesky OAuth flow can authenticate directly against your PDS without extra auth-server infrastructure.
  • The built-in provider publishes /.well-known/oauth-protected-resource, /.well-known/oauth-authorization-server, /oauth/jwks, /oauth/par, /oauth/authorize, /oauth/token, and /oauth/revoke from the same host as the PDS.
  • OAuth scope enforcement now understands both the transition scopes (transition:generic, transition:email, transition:chat.bsky), the newer granular permission families (account:, identity:, repo:, blob:, and rpc:), and include:<nsid> permission-set scopes, so clients that request narrower ATProto permissions can be authorized without silently getting broader access.
  • If service_handle_domain is example.com, submitting handle: "alice" to com.atproto.server.createAccount creates alice.example.com.
  • If invite_code_required is enabled, public signup is disabled until a valid invite code is supplied.
  • com.atproto.server.createInviteCode and com.atproto.server.createInviteCodes are admin-only by default. Set self_service_invite_codes to enable self-service invite minting for authenticated full-access sessions, limited to the caller's own account.
  • script/perlsky-admin create-invite can mint invite codes locally on the server without needing an existing user session.
  • The invite-only bootstrap flow is documented with copy-pasteable commands in docs/DEPLOYMENT.md.
  • Browser clients such as bsky.app can talk to perlsky directly because XRPC and DID-document responses include CORS headers and answer OPTIONS preflight requests.
  • OAuth clients such as Tangled can also discover and use perlsky directly as both the protected resource and authorization server, using PAR, PKCE, private_key_jwt, and DPoP as required by the ATProto OAuth profile.
  • Unknown app.bsky.* requests are proxied to https://api.bsky.app by default, and unknown chat.bsky.* requests are proxied to https://api.bsky.chat by default using per-account service-auth JWTs.
  • Set bsky_appview_url / bsky_appview_did or chat_service_url / chat_service_did in your config if you want different upstream services.

Relay / crawler discovery:

  • Configure hostname to the public host name you want relays to crawl, for example pds.example.com. This should be the host, not the full URL.
  • Configure crawlers as a list of relay or crawler service origins, for example ["https://bsky.network"].
  • perlsky will POST com.atproto.sync.requestCrawl to each configured crawler after local repo/account/identity activity, while throttling repeat notices with crawler_notify_interval (default 1200 seconds).
  • Local regression coverage for this path lives in t/crawlers.t.

Browser smoke:

  • script/perlsky-browser-smoke run-dual now prefers a saved reusable account pair from .cache/browser-smoke/reusable-pair.json, so the default dual-account smoke does not create new actors.
  • Use script/perlsky-browser-smoke bootstrap-pair once to mint and save a dedicated smoke pair, then rerun script/perlsky-browser-smoke run-dual as often as you want against those same accounts.
  • Use script/perlsky-browser-smoke show-pair to inspect the saved pair metadata and script/perlsky-browser-smoke clear-pair to forget it locally.
  • The reusable dual-account smoke now covers list lifecycle plus deeper bsky.app settings flows, and it pre-cleans old smoke-created posts/lists before each run.
  • DMs are intentionally deferred from the current browser-smoke tranche; see docs/BROWSER_SMOKE.md for current scope and rationale.
  • Fresh-account creation is still available through the explicit bootstrap-* commands, but it is no longer the normal path for repeated browser smoke runs.
  • Detailed browser-smoke workflow, current interaction coverage, and the env-gated prove wrapper live in docs/BROWSER_SMOKE.md.
  • Extraction work toward a cross-PDS standalone package now lives in the atproto-smoke repo, which owns the browser runtime, package CLI, example configs, and bring-your-own-account plus perlsky adapter helpers.
  • script/perlsky-browser-smoke expects a standalone checkout at ../atproto-smoke by default. Set PERLSKY_BROWSER_SUITE_ROOT to point it at any other checkout explicitly.
  • The lowest-friction local layout is to clone atproto-smoke next to perlsky, so both repos share the same parent directory.
  • The shared smoke runtime now applies a bounded per-step timeout (stepTimeoutMs, default 120000) so late browser stalls fail with artifacts instead of hanging forever.

Moderation and labels:

  • com.atproto.admin.updateSubjectStatus now enforces repo, record, and blob takedowns as real behavior instead of passive metadata.
  • Repo takedowns block ordinary login, repo writes, and public repo reads. allowTakendown sessions are accepted for parity with the reference PDS, but those sessions still cannot write.
  • Record takedowns hide records from com.atproto.repo.getRecord and com.atproto.repo.listRecords.
  • Blob takedowns quarantine blob reads for the public while still permitting authenticated self/admin recovery access, and they block both duplicate blob uploads and new record writes that reference quarantined blobs.
  • com.atproto.label.queryLabels, com.atproto.label.subscribeLabels, and com.atproto.temp.fetchLabels are backed by persisted local labels rather than synthesized snapshots. Admin takedowns emit !hide labels and restores emit negation events.
  • Label query/stream behavior is covered by local regression tests in t/labels.t. The official reference PDS does not provide a like-for-like local labeler implementation to diff against, so direct upstream differential checks are focused on moderation semantics rather than label RPC parity.

Interop fixtures:

  • t/crypto-interop.t loads the official Bluesky tools/reference/atproto/interop-test-files/crypto/w3c_didkey_K256.json vectors so secp256k1 did:key encoding stays pinned to the same public fixtures as the upstream stack.
  • t/plc-identity.t drives perlsky against the local PLC mock built on the official @did-plc/lib, covering account creation, recommended DID credentials, PLC handle updates, token-gated PLC signing, and validated PLC submission semantics.