Files
indiekit-endpoint-activitypub/CLAUDE.md
Ricardo ab2363d123 docs: update CLAUDE.md and README.md for v2.5.0
Document unified item processing pipeline (gotcha #23), parameterized
infinite scroll component (gotcha #24), quote embeds (gotcha #25).
Update architecture tree with new modules and controllers. Expand
route table and admin UI pages with explore, tag timeline, post detail,
and API endpoints. Add reader features (explore, hashtags, quotes,
link previews, read tracking, infinite scroll) to README.

Confab-Link: http://localhost:8080/sessions/e9d666ac-3c90-4298-9e92-9ac9d142bc06
2026-03-03 13:03:48 +01:00

25 KiB

CLAUDE.md — @rmdes/indiekit-endpoint-activitypub

AI agent instructions for working on this codebase. Read this entire file before making any changes.

What This Is

An Indiekit plugin that adds full ActivityPub federation via Fedify. It turns an Indiekit-powered IndieWeb site into a fediverse actor — discoverable, followable, and interactive from Mastodon, Misskey, Pixelfed, Lemmy, etc.

npm: @rmdes/indiekit-endpoint-activitypub Version: See package.json Node: >=22 Module system: ESM ("type": "module")

Architecture Overview

index.js                          ← Plugin entry, route registration, syndicator
├── lib/federation-setup.js       ← Fedify Federation instance, dispatchers, collections
├── lib/federation-bridge.js      ← Express ↔ Fedify request/response bridge
├── lib/inbox-listeners.js        ← Handlers for Follow, Undo, Like, Announce, Create, Delete, etc.
├── lib/jf2-to-as2.js             ← JF2 → ActivityStreams conversion (plain JSON + Fedify vocab)
├── lib/kv-store.js               ← MongoDB-backed KvStore for Fedify (get/set/delete/list)
├── lib/activity-log.js           ← Activity logging to ap_activities
├── lib/item-processing.js        ← Unified item processing pipeline (moderation, quotes, interactions, rendering)
├── lib/timeline-store.js         ← Timeline item extraction + sanitization
├── lib/timeline-cleanup.js       ← Retention-based timeline pruning
├── lib/og-unfurl.js              ← Open Graph link previews + quote enrichment
├── lib/batch-refollow.js         ← Gradual re-follow for imported Mastodon accounts
├── lib/migration.js              ← CSV parsing + WebFinger resolution for Mastodon import
├── lib/csrf.js                   ← CSRF token generation/validation
├── lib/storage/
│   ├── timeline.js               ← Timeline CRUD with cursor pagination
│   ├── notifications.js          ← Notification CRUD with read/unread tracking
│   └── moderation.js             ← Mute/block storage
├── lib/controllers/              ← Express route handlers (admin UI)
│   ├── dashboard.js, reader.js, compose.js, profile.js, profile.remote.js
│   ├── public-profile.js         ← Public profile page (HTML fallback for actor URL)
│   ├── explore.js, explore-utils.js ← Explore public Mastodon timelines
│   ├── hashtag-explore.js        ← Cross-instance hashtag search
│   ├── tag-timeline.js           ← Posts filtered by hashtag
│   ├── post-detail.js            ← Single post detail view
│   ├── api-timeline.js           ← AJAX API for infinite scroll + new post count
│   ├── followers.js, following.js, activities.js
│   ├── featured.js, featured-tags.js
│   ├── interactions.js, interactions-like.js, interactions-boost.js
│   ├── moderation.js, migrate.js, refollow.js
├── views/                        ← Nunjucks templates
│   ├── activitypub-*.njk         ← Page templates
│   ├── layouts/ap-reader.njk     ← Reader layout (NOT reader.njk — see gotcha below)
│   └── partials/                 ← Shared components (item card, quote embed, link preview, media)
├── assets/
│   ├── reader.css                ← Reader UI styles
│   ├── reader-infinite-scroll.js ← Alpine.js components (infinite scroll, new posts banner, read tracking)
│   ├── reader-tabs.js            ← Alpine.js tab persistence
│   └── icon.svg                  ← Plugin icon
└── locales/en.json               ← i18n strings

Data Flow

Outbound: Indiekit post → syndicator.syndicate() → jf2ToAS2Activity() → ctx.sendActivity() → follower inboxes
Inbound:  Remote inbox POST → Fedify → inbox-listeners.js → MongoDB collections → admin UI
Reader:   Followed account posts → Create inbox → timeline-store → ap_timeline → reader UI
Explore:  Public Mastodon API → fetchMastodonTimeline() → mapMastodonToItem() → explore UI

All views (reader, explore, tag timeline, hashtag explore, API endpoints) share a single
processing pipeline via item-processing.js:
  items → applyTabFilter() → loadModerationData() → postProcessItems() → render

MongoDB Collections

Collection Purpose Key fields
ap_followers Accounts following us actorUrl (unique), inbox, sharedInbox, source
ap_following Accounts we follow actorUrl (unique), source, acceptedAt
ap_activities Activity log (TTL-indexed) direction, type, actorUrl, objectUrl, receivedAt
ap_keys Cryptographic key pairs type ("rsa" or "ed25519"), key material
ap_kv Fedify KvStore + job state _id (key path), value
ap_profile Actor profile (single doc) name, summary, icon, attachments, actorType
ap_featured Pinned posts postUrl, pinnedAt
ap_featured_tags Featured hashtags tag, addedAt
ap_timeline Reader timeline items uid (unique), published, author, content
ap_notifications Likes, boosts, follows, mentions uid (unique), type, read
ap_muted Muted actors/keywords url or keyword
ap_blocked Blocked actors url
ap_interactions Like/boost tracking per post objectUrl, type

Critical Patterns and Gotchas

1. Express ↔ Fedify Bridge (CUSTOM — NOT @fedify/express)

We cannot use @fedify/express's integrateFederation() because Indiekit mounts plugins at sub-paths. Express strips the mount prefix from req.url, breaking Fedify's URI template matching. Verified in Fedify 2.0: @fedify/express still uses req.url (not req.originalUrl), so the custom bridge remains necessary. Instead, federation-bridge.js uses req.originalUrl to build the full URL.

The bridge also reconstructs POST bodies from req.body when Express body parser has already consumed the request stream (checked via req.readable === false). Without this, POST handlers in Fedify (e.g. the @fedify/debugger login form) receive empty bodies and fail with "Response body object should not be disturbed or locked".

If you see path-matching issues with Fedify, check that req.originalUrl is being used, not req.url.

2. Content Negotiation Route — GET Only

The contentNegotiationRoutes router is mounted at / (root). It MUST only pass GET/HEAD requests to Fedify. Passing POST/PUT/DELETE would cause fromExpressRequest() to consume the body stream via Readable.toWeb(req), breaking Express body-parsed routes downstream (admin forms, Micropub, etc.).

3. Skip Fedify for Admin Routes

In routesPublic, the middleware skips paths starting with /admin. Without this, Fedify would intercept admin UI requests and return 404/406 responses instead of letting Express serve the authenticated pages.

4. Authenticated Document Loader for Inbox Handlers

All .getObject() / .getActor() / .getTarget() calls in inbox handlers must pass an authenticated DocumentLoader to sign outbound fetches. Without this, requests to Authorized Fetch (Secure Mode) servers like hachyderm.io fail with 401.

const authLoader = await ctx.getDocumentLoader({ identifier: handle });
const actor = await activity.getActor({ documentLoader: authLoader });
const object = await activity.getObject({ documentLoader: authLoader });

The getAuthLoader helper in inbox-listeners.js wraps this pattern. The authenticated loader is also passed through to extractObjectData() and extractActorInfo() in timeline-store.js so that .getAttributedTo(), .getIcon(), .getTags(), and .getAttachments() also sign their fetches.

Still prefer .objectId?.href and .actorId?.href (zero network requests) when you only need the URL — e.g. Like, Delete, and the filter check in Announce. Only use the fetching getters when you need the full object, and always wrap in try-catch.

5. Accept(Follow) Matching — Don't Check Inner Object Type

Fedify often resolves the inner object of Accept to a Person (the Follow's target) rather than the Follow itself. The Accept handler matches against ap_following by actor URL instead of inspecting inner instanceof Follow.

6. Filter Inbound Likes/Announces to Our Content Only

Without filtering, the inbox logs every Like/Announce from every federated server — including reactions to other people's content that happens to flow through shared inboxes. Check objectId.startsWith(publicationUrl) before logging.

7. Nunjucks Template Name Collisions

Template names resolve across ALL registered plugin view directories. If two plugins have views/layouts/reader.njk, Nunjucks loads whichever it finds first (often wrong). The reader layout is named ap-reader.njk to avoid collision with @rmdes/indiekit-endpoint-microsub's reader.njk.

Never name a layout/template with a generic name that another plugin might use.

8. Express 5 — No redirect("back")

Express 5 removed the "back" magic keyword from response.redirect(). It's treated as a literal URL, causing 404s at paths like /admin/featured/back. Always use explicit redirect paths.

9. Attachment Array Workaround (Mastodon Compatibility)

JSON-LD compaction collapses single-element arrays to plain objects. Mastodon's update_account_fields checks attachment.is_a?(Array) and silently skips if it's not an array. sendFedifyResponse() in federation-bridge.js forces attachment to always be an array.

Note: The old endpoints.type bug (fedify#576) was fixed in Fedify 2.0 — that workaround has been removed.

express.urlencoded({ extended: true }) uses qs which strips [] from array field names. HTML fields named link_name[] arrive as request.body.link_name (not request.body["link_name[]"]). The profile controller reads link_name and link_value, NOT link_name[].

11. Author Resolution Fallback Chain

extractObjectData() in timeline-store.js uses a multi-strategy fallback:

  1. object.getAttributedTo() — async, may fail with Authorized Fetch
  2. options.actorFallback — the activity's actor (passed from Create handler)
  3. object.attribution / object.attributedTo — plain object properties
  4. object.attributionIds — non-fetching URL array with username extraction from common patterns (/@name, /users/name)

Without this chain, many timeline items show "Unknown" as the author.

12. Username Extraction from Actor URLs

When extracting usernames from attribution IDs, handle multiple URL patterns:

  • /@username (Mastodon)
  • /users/username (Mastodon, Indiekit)
  • /ap/users/12345/ (numeric IDs on some platforms)

The regex was previously matching "users" instead of the actual username from /users/NatalieDavis.

13. Empty Boost Filtering

Lemmy/PieFed send Announce activities where the boosted object resolves to an activity ID instead of a Note/Article with actual content. Check object.content || object.name before storing to avoid empty cards in the timeline.

14. Temporal.Instant for Fedify Dates

Fedify uses @js-temporal/polyfill for dates. When setting published on Fedify objects, use Temporal.Instant.from(isoString). When reading Fedify dates in inbox handlers, use String(object.published) to get ISO strings — NOT new Date(object.published) which causes TypeError.

15. LogTape — Configure Once Only

@logtape/logtape's configure() can only be called once per process. The module-level _logtapeConfigured flag prevents duplicate configuration. If configure fails (e.g., another plugin already configured it), catch the error silently.

When the debug dashboard is enabled (debugDashboard: true), LogTape configuration is skipped entirely because @fedify/debugger configures its own LogTape sink for the dashboard UI.

16. .authorize() Intentionally NOT Chained on Actor Dispatcher

Fedify's .authorize() triggers HTTP Signature verification on every GET to the actor endpoint. Servers requiring Authorized Fetch cause infinite loops: Fedify tries to fetch their key → they return 401 → Fedify retries → 500 errors. Re-enable when Fedify supports authenticated document loading for outgoing fetches.

17. Delivery Queue Must Be Started

federation.startQueue() MUST be called after setup. Without it, ctx.sendActivity() enqueues tasks but the message queue never processes them — activities are never delivered.

18. Shared Key Dispatcher for Shared Inbox

inboxChain.setSharedKeyDispatcher() tells Fedify to use our actor's key pair when verifying HTTP Signatures on the shared inbox. Without this, servers like hachyderm.io (which requires Authorized Fetch) have their signatures rejected.

19. Fedify 2.0 Modular Imports

Fedify 2.0 uses modular entry points instead of a single barrel export. Imports must use the correct subpath:

// Core federation infra
import { createFederation, InProcessMessageQueue } from "@fedify/fedify";

// Crypto operations (key generation, import/export)
import { exportJwk, generateCryptoKeyPair, importJwk } from "@fedify/fedify/sig";

// ActivityStreams vocabulary types
import { Person, Note, Article, Create, Follow, ... } from "@fedify/fedify/vocab";

// WRONG (Fedify 1.x style) — these no longer work:
// import { Person, createFederation, exportJwk } from "@fedify/fedify";

20. importSpki Removed in Fedify 2.0

Fedify 1.x exported importSpki() for loading PEM public keys. This was removed in 2.0. The local importSpkiPem() function in federation-setup.js replaces it using the Web Crypto API directly (crypto.subtle.importKey("spki", ...)). Similarly, importPkcs8Pem() handles private keys in PKCS#8 format.

21. KvStore Requires list() in Fedify 2.0

Fedify 2.0 added a list(prefix?) method to the KvStore interface. It must return an AsyncIterable<{ key: string[], value: unknown }>. The MongoKvStore in kv-store.js implements this as an async generator that queries MongoDB with a regex prefix match on the _id field.

22. Debug Dashboard Body Consumption

The @fedify/debugger login form POSTs application/x-www-form-urlencoded data. Because Express's body parser runs before the Fedify bridge, the POST body stream is already consumed (req.readable === false). The bridge in federation-bridge.js detects this and reconstructs the body from req.body. Without this, the debugger's login handler receives an empty body and throws "Response body object should not be disturbed or locked". See also Gotcha #1.

23. Unified Item Processing Pipeline

All views that display timeline items — reader, explore, tag timeline, hashtag explore, and their AJAX API counterparts — must use the shared pipeline in lib/item-processing.js. Never duplicate moderation filtering, quote stripping, interaction map building, or card rendering in individual controllers.

The pipeline flow is:

import { postProcessItems, applyTabFilter, loadModerationData, renderItemCards } from "../item-processing.js";

// 1. Get raw items (from MongoDB or Mastodon API)
// 2. Filter by tab/type (optional)
const filtered = applyTabFilter(items, tab);
// 3. Load moderation data once
const moderation = await loadModerationData(modCollections);
// 4. Run unified pipeline (filters muted/blocked, strips quote refs, builds interaction map)
const { items: processed, interactionMap } = await postProcessItems(filtered, { moderation, interactionsCol });
// 5. For AJAX endpoints, render HTML server-side
const html = await renderItemCards(processed, request, { interactionMap, mountPath, csrfToken });

Key functions:

  • postProcessItems() — orchestrates moderation → quote stripping → interaction map
  • applyModerationFilters() — filters items by muted URLs, keywords, blocked URLs
  • stripQuoteReferences() — removes inline RE: <link> paragraphs when quote embed exists
  • buildInteractionMap() — queries ap_interactions for like/boost state per item
  • applyTabFilter() — filters items by type tab (notes, articles, replies, boosts, media)
  • renderItemCards() — server-side Nunjucks rendering of ap-item-card.njk for AJAX responses
  • loadModerationData() — convenience wrapper to load muted/blocked data from MongoDB

If you add a new view that shows timeline items, use this pipeline. Do not inline the logic.

24. Unified Infinite Scroll Alpine Component

All views with infinite scroll use a single apInfiniteScroll Alpine.js component (in assets/reader-infinite-scroll.js), parameterized via data attributes on the container element:

<div class="ap-load-more"
  data-cursor="{{ cursor }}"
  data-api-url="{{ mountPath }}/admin/reader/api/timeline"
  data-cursor-param="before"        <!-- query param name sent to API -->
  data-cursor-field="before"         <!-- response JSON field for next cursor -->
  data-timeline-id="ap-timeline"     <!-- DOM ID to append HTML into -->
  data-extra-params='{{ extraJson }}'  <!-- JSON object of additional query params -->
  data-hide-pagination="pagination-id" <!-- optional: ID of no-JS pagination to hide -->
  x-data="apInfiniteScroll()"
  x-init="init()">

Do not create separate scroll components for new views. Configure the existing one with appropriate data attributes. The explore view uses data-cursor-param="max_id" and data-cursor-field="maxId" (Mastodon API conventions), while the reader uses data-cursor-param="before" and data-cursor-field="before".

25. Quote Embeds and Enrichment

Posts that quote another post (Mastodon quote feature via FEP-044f) are rendered with an embedded card showing the quoted post's author, content, and timestamp. The data flow:

  1. Ingest: extractObjectData() reads object.quoteUrl (Fedify reads as:quoteUrl, misskey:_misskey_quote, fedibird:quoteUri)
  2. Enrichment: fetchAndStoreQuote() in og-unfurl.js fetches the quoted post via ctx.lookupObject(), extracts data with extractObjectData(), and stores it as quote on the timeline item
  3. On-demand: post-detail.js fetches quotes on demand for items that have quoteUrl but no stored quote data (pre-existing items)
  4. Rendering: partials/ap-quote-embed.njk renders the embedded card; stripQuoteReferences() removes the inline RE: <link> paragraph to avoid duplication

Date Handling Convention

All dates MUST be stored as ISO 8601 strings. This is mandatory across all Indiekit plugins.

// CORRECT
followedAt: new Date().toISOString()
published: String(fedifyObject.published)  // Temporal → string

// WRONG — crashes Nunjucks | date filter
followedAt: new Date()
published: new Date(fedifyObject.published)

The Nunjucks | date filter calls date-fns parseISO() which only accepts ISO strings. Date objects cause "dateString.split is not a function" crashes.

Batch Re-follow State Machine

import → refollow:pending → refollow:sent → federation  (happy path: Accept received)
import → refollow:pending → refollow:sent → refollow:failed (after 3 retries)
  • import: Imported from Mastodon CSV, no Follow sent yet
  • refollow:pending: Claimed by batch processor, being processed
  • refollow:sent: Follow activity sent, awaiting Accept
  • federation: Accept received, fully federated
  • refollow:failed: Max retries exceeded

On restart, refollow:pending entries are reset to import to prevent stale claims.

Plugin Lifecycle

  1. constructor() — Merges options with defaults
  2. init(Indiekit) — Called by Indiekit during startup:
    • Stores publication.me as _publicationUrl
    • Registers 13 MongoDB collections with indexes
    • Seeds actor profile from config (first run only)
    • Calls setupFederation() which creates Fedify instance + starts queue
    • Registers endpoint (mounts routes) and syndicator
    • Starts batch re-follow processor (10s delay)
    • Schedules timeline cleanup (on startup + every 24h)

Route Structure

Method Path Handler Auth
* /.well-known/* Fedify (WebFinger, NodeInfo) No
* {mount}/users/*, {mount}/inbox Fedify (actor, inbox, outbox, collections) No (HTTP Sig)
GET {mount}/ Dashboard Yes (IndieAuth)
GET {mount}/admin/reader Timeline reader Yes
GET {mount}/admin/reader/explore Explore public Mastodon timelines Yes
GET {mount}/admin/reader/explore/hashtag Cross-instance hashtag search Yes
GET {mount}/admin/reader/tag Tag timeline (posts by hashtag) Yes
GET {mount}/admin/reader/post Post detail view Yes
GET {mount}/admin/reader/notifications Notifications Yes
GET {mount}/admin/reader/api/timeline AJAX timeline API (infinite scroll) Yes
GET {mount}/admin/reader/api/timeline/count-new New post count API (polling) Yes
POST {mount}/admin/reader/api/timeline/mark-read Mark posts as read API Yes
GET {mount}/admin/reader/api/explore AJAX explore API (infinite scroll) Yes
POST {mount}/admin/reader/compose Compose reply Yes
POST {mount}/admin/reader/like,unlike,boost,unboost Interactions Yes
POST {mount}/admin/reader/follow,unfollow Follow/unfollow Yes
POST {mount}/admin/reader/follow-tag,unfollow-tag Follow/unfollow hashtag Yes
GET {mount}/admin/reader/profile Remote profile view Yes
GET {mount}/admin/reader/moderation Moderation dashboard Yes
POST {mount}/admin/reader/mute,unmute,block,unblock Moderation actions Yes
GET {mount}/admin/followers,following,activities Lists Yes
GET/POST {mount}/admin/profile Actor profile editor Yes
GET/POST {mount}/admin/featured Pinned posts Yes
GET/POST {mount}/admin/tags Featured tags Yes
GET/POST {mount}/admin/migrate Mastodon migration Yes
* {mount}/admin/refollow/* Batch refollow control Yes
* {mount}/__debug__/* Fedify debug dashboard (if enabled) Password
GET {mount}/users/:identifier Public profile page (HTML fallback) No
GET /* (root) Content negotiation (AP clients only) No

Dependencies

Package Purpose
@fedify/fedify ActivityPub federation framework (v2.0+)
@fedify/debugger Optional debug dashboard with OpenTelemetry tracing
@fedify/redis Redis message queue for delivery
@js-temporal/polyfill Temporal API for Fedify date handling
ioredis Redis client
sanitize-html XSS prevention for timeline/notification content
unfurl.js Open Graph metadata extraction for link previews
express Route handling (peer: Indiekit provides it)

Configuration Options

{
  mountPath: "/activitypub",         // URL prefix for all routes
  actor: {
    handle: "rick",                  // Fediverse username
    name: "Ricardo Mendes",          // Display name (seeds profile)
    summary: "",                     // Bio (seeds profile)
    icon: "",                        // Avatar URL (seeds profile)
  },
  checked: true,                     // Syndicator checked by default
  alsoKnownAs: "",                   // Mastodon migration alias
  activityRetentionDays: 90,         // TTL for ap_activities (0 = forever)
  storeRawActivities: false,         // Store full JSON of inbound activities
  redisUrl: "",                      // Redis for delivery queue (empty = in-process)
  parallelWorkers: 5,               // Parallel delivery workers (with Redis)
  actorType: "Person",              // Person | Service | Organization | Group
  logLevel: "warning",             // Fedify log level: debug | info | warning | error | fatal
  timelineRetention: 1000,          // Max timeline items (0 = unlimited)
  notificationRetentionDays: 30,    // Days to keep notifications (0 = forever)
  debugDashboard: false,            // Enable @fedify/debugger dashboard at {mount}/__debug__/
  debugPassword: "",                // Password for debug dashboard (required if dashboard enabled)
}

Publishing Workflow

  1. Edit code in this repo
  2. Bump version in package.json (npm rejects duplicate versions)
  3. Commit and push
  4. STOP — user must run npm publish manually (requires OTP)
  5. After publish confirmation, update Dockerfile version in indiekit-cloudron/
  6. cloudron build --no-cache && cloudron update --app rmendes.net --no-backup

Testing

No automated test suite. Manual testing against real fediverse servers:

# WebFinger
curl -s "https://rmendes.net/.well-known/webfinger?resource=acct:rick@rmendes.net" | jq .

# Actor document
curl -s -H "Accept: application/activity+json" "https://rmendes.net/" | jq .

# NodeInfo
curl -s "https://rmendes.net/nodeinfo/2.1" | jq .

# Search from Mastodon for @rick@rmendes.net

CSS Conventions

The reader CSS (assets/reader.css) uses Indiekit's theme custom properties for automatic dark mode support:

  • --color-on-background (not --color-text)
  • --color-on-offset (not --color-text-muted)
  • --border-radius-small (not --border-radius)
  • --color-red45, --color-green50, etc. (not hardcoded hex)

Post types are differentiated by left border color: purple (notes), green (articles), yellow (boosts), primary (replies).