sifa-rss-import #

Import a site's articles (via its RSS/Atom feeds) into the Publications section of a Sifa ID profile, with deduplication of what already exists.

Under the hood: Sifa is an AppView that reads the records from a user's atproto PDS. This tool therefore writes id.sifa.profile.publication records straight into your repo — they then show up automatically on your Sifa profile.

How it works #

1. Discovery — reads the <link rel="alternate" type="application/rss+xml"> tags from the site (fallback: /index.xml, /feed, … for Hugo & co).
2. Parsing — RSS 2.0 + Atom (rss-parser).
3. Deduplication — lists the publications already present (com.atproto.repo.listRecords) and compares by normalized URL.
4. Writing — creates only the missing ones (com.atproto.repo.createRecord), with a concurrency limit and retries.

RSS → id.sifa.profile.publication mapping:

RSS/Atom	Sifa field
<title>	title
<link>	url
feed title	publisher
<description>	description
<pubDate>	publishedAt (YYYY-MM-DD)
—	createdAt (now)

Requirements #

• Node.js ≥ 20
• An account on an atproto PDS (your Sifa/Bluesky handle)

Install #

npm install

Authentication (two modes) #

Dev loopback mode — default, no hosting required. The OAuth client_id is a loopback client (http://localhost?redirect_uri=...&scope=...): the authorization server recognizes it without fetching any URL. This is the ideal mode for personal use from your machine — nothing to drop on your site.

Hosted mode — for automation. atproto then requires the client_id to be a public HTTPS URL serving client-metadata.json. See Automation below.

1. Authorize (once) #

npm run login -- une-tasse-de.cafe

A browser opens for authorization; the callback comes back to http://127.0.0.1:8765/callback. The session (refresh token + DPoP key) is persisted in ~/.config/sifa-rss-import/sessions.json — subsequent runs refresh without a browser. The command prints your DID to reuse.

2. Import #

# Preview without writing anything:
npm run import -- https://une-tasse-de.cafe --handle did:plc:hqnyog7skad6m4aejb2yujxy --dry-run

# For real:
npm run import -- https://une-tasse-de.cafe --handle did:plc:hqnyog7skad6m4aejb2yujxy

Options: --dry-run (writes nothing), --limit <n> (caps the number of creations), --include <s> / --exclude <s> (repeatable URL filters, see below). Re-running is idempotent: only new entries are created.

Filtering out non-article pages #

A site's root feed often aggregates standalone pages (an "about" page, a CV, an archive index…) alongside real articles. Restrict the import with --include (keep only URLs containing the substring) and/or --exclude (drop matching URLs); both are repeatable and match anywhere in the URL:

# Keep only /blog/ and /expresso/ articles:
npm run import -- https://une-tasse-de.cafe --handle did:plc:hqnyog7skad6m4aejb2yujxy \
  --include /blog/ --include /expresso/

# Or drop specific pages:
npm run import -- https://une-tasse-de.cafe --handle did:plc:hqnyog7skad6m4aejb2yujxy \
  --exclude /cv/ --exclude /talks/

Scope #

The default requested scope is atproto transition:generic (works everywhere, covers repo writes). The clean, granular scope is the id.sifa.authProfile permission set (it grants exactly id.sifa.profile.* + id.sifa.graph.follow); its support on atproto authorization servers is still rolling out. To try it:

SIFA_OAUTH_SCOPE="atproto include:id.sifa.authProfile" npm run login -- une-tasse-de.cafe

Automation (Tangled) #

For a recurring, non-interactive import, switch to hosted mode:

1. Host client-metadata.json at a public HTTPS URL — on your Hugo site, drop it into static/ and deploy (→ https://une-tasse-de.cafe/client-metadata.json). The file's client_id must match that URL exactly.
2. Set SIFA_CLIENT_ID=https://une-tasse-de.cafe/client-metadata.json: this enables hosted mode (loads client-metadata.json).
3. Run login once in this mode to obtain a hosted session.

Since the refresh token is persisted, subsequent runs work without a browser. On Tangled, use a pipeline (not a GitHub Action): inject the contents of sessions.json as a secret (with SIFA_DATA_DIR pointing to a directory rebuilt at run time) and run npm run import -- <site> --handle <did>.

License #

MIT