@sockethub/platform-metadata 1.0.1-alpha.9 → 1.0.1-beta.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -77,9 +77,72 @@ The platform returns an ActivityStream object with extracted metadata:
77
77
  The platform extracts the following metadata when available:
78
78
 
79
79
  * **Open Graph data**: Title, description, images, site name, URL, locale
80
- * **Site information**: Favicon, character encoding
80
+ * **Site information**: Favicon, character encoding — when a page declares no
81
+ icon, `favicon` falls back to the conventional relative `/favicon.ico` so
82
+ clients can resolve it against the page URL and decorate previews
81
83
  * **Canonical URLs**: Resolved and canonical page URLs
82
- * **Media**: Images with dimensions and type information
84
+ * **Media**: Images with dimensions and type information; for X/Twitter posts,
85
+ a `video` object (direct media URL, poster `thumbnail`, dimensions, duration)
86
+
87
+ ## Site-Specific Handling
88
+
89
+ Some large platforms don't expose a post's own media to Open Graph scrapers,
90
+ so those URLs are resolved through preview services built for this purpose:
91
+
92
+ * **X / Twitter status URLs** (`x.com`, `twitter.com`, incl. `/i/web/status/…`)
93
+ are fetched from the [FxTwitter API](https://github.com/FxEmbed/FxEmbed)
94
+ instead of scraped — X serves only a generic site banner to scrapers. The
95
+ response carries the post's text, its actual photo (or video + poster
96
+ thumbnail), and the canonical `x.com` URL. If the API is unavailable or the
97
+ post can't be resolved, the platform falls back to the regular scrape
98
+ (which still yields the post text).
99
+ * **Reddit post URLs** use `old.reddit.com`'s structured JSON response, avoiding
100
+ Reddit's unreliable HTML and Open Graph crawler behavior. The post title,
101
+ self-text, and direct `i.redd.it` image are mapped into the preview; text posts
102
+ intentionally have no image. oEmbed remains a title-only failure fallback.
103
+ See `compatUserAgent` below.
104
+ * **Facebook** remains best-effort: post text usually comes through. The
105
+ video or photo thumbnail is taken from the post's Open Graph image, then
106
+ its Twitter card image, and finally the image rendered by Facebook's public
107
+ embed plugin (`/plugins/video.php` or `/plugins/post.php`), which Facebook
108
+ serves to third parties when it withholds Open Graph media from a
109
+ datacenter address. Full media still requires authentication.
110
+
111
+ ### `userAgent`
112
+
113
+ All outbound requests identify as
114
+ `Mozilla/5.0 (compatible; SockethubBot/<version>; +https://sockethub.org)`.
115
+ Sites gate scraper-facing behavior on the user agent — preview services serve
116
+ their Open Graph payload to bot-like UAs, and many hosts reject undici's
117
+ default UA. Override per deployment via `packageConfig`:
118
+
119
+ ```json
120
+ {
121
+ "packageConfig": {
122
+ "@sockethub/platform-metadata": {
123
+ "userAgent": "MyDeployment/1.0 (+https://my.example)"
124
+ }
125
+ }
126
+ }
127
+ ```
128
+
129
+ ### `compatUserAgent`
130
+
131
+ Sites that only serve metadata to *recognized* embed crawlers (Reddit)
132
+ are fetched with a link-preview crawler user agent instead — the established
133
+ practice for self-hosted preview fetchers. Defaults to
134
+ `Mozilla/5.0 (compatible; Discordbot/2.0; +https://discordapp.com)`;
135
+ override via `packageConfig`:
136
+
137
+ ```json
138
+ {
139
+ "packageConfig": {
140
+ "@sockethub/platform-metadata": {
141
+ "compatUserAgent": "Mozilla/5.0 (compatible; TelegramBot/1.0)"
142
+ }
143
+ }
144
+ }
145
+ ```
83
146
 
84
147
  ## Dependencies
85
148
 
@@ -92,9 +155,14 @@ Because the server fetches whatever URL a client supplies as `actor.id`, metadat
92
155
  requests are hardened against server-side request forgery (SSRF) and resource
93
156
  exhaustion. The fetch runs through a guarded connection layer that:
94
157
 
158
+ * rejects URL schemes other than `http` and `https`;
95
159
  * refuses to connect to loopback, private, link-local, carrier-grade NAT, or
96
- cloud-metadata addresses (e.g. `169.254.169.254`), in every IPv4/IPv6 spelling,
97
- validated at connection time so it also covers redirect hops;
160
+ cloud-metadata addresses (e.g. `169.254.169.254`), validated at connection
161
+ time so it also covers redirect hops. IPv4-mapped, IPv4-compatible, and
162
+ well-known NAT64 (`64:ff9b::/96`) embeddings are judged as the embedded IPv4
163
+ address. The local-use NAT64 prefix `64:ff9b:1::/48` is refused entirely.
164
+ The ranges are listed under
165
+ [Outbound URL guard](../../docs/configuration.md#outbound-url-guard);
98
166
  * caps the response body size.
99
167
 
100
168
  ### `allowPrivateAddresses`
package/dist/index.d.ts CHANGED
@@ -1,6 +1,9 @@
1
1
  import type { ActivityStream, PlatformCallback, PlatformConfig, PlatformInterface, PlatformSession } from "@sockethub/schemas";
2
+ /** Enforce a deadline independently of a dependency's AbortSignal handling. */
3
+ export declare function withDeadline<T>(promise: Promise<T>, timeoutMs: number): Promise<T>;
2
4
  export default class Metadata implements PlatformInterface {
3
5
  private readonly log;
6
+ private readonly fetchImpl;
4
7
  private dispatcher?;
5
8
  config: PlatformConfig;
6
9
  constructor(session: PlatformSession);
@@ -88,6 +91,28 @@ export default class Metadata implements PlatformInterface {
88
91
  image: {
89
92
  $ref: string;
90
93
  };
94
+ video: {
95
+ type: string;
96
+ required: string[];
97
+ additionalProperties: boolean;
98
+ properties: {
99
+ url: {
100
+ type: string;
101
+ };
102
+ thumbnail: {
103
+ type: string;
104
+ };
105
+ width: {
106
+ type: string;
107
+ };
108
+ height: {
109
+ type: string;
110
+ };
111
+ duration: {
112
+ type: string;
113
+ };
114
+ };
115
+ };
91
116
  url: {
92
117
  type: string;
93
118
  };
@@ -118,6 +143,9 @@ export default class Metadata implements PlatformInterface {
118
143
  height: {
119
144
  type: string[];
120
145
  };
146
+ alt: {
147
+ type: string;
148
+ };
121
149
  };
122
150
  };
123
151
  };
@@ -129,6 +157,33 @@ export default class Metadata implements PlatformInterface {
129
157
  * Stateless platforms are always ready to handle jobs.
130
158
  */
131
159
  isInitialized(): boolean;
160
+ private userAgent;
161
+ private compatUserAgent;
132
162
  fetch(job: ActivityStream, cb: PlatformCallback): void;
163
+ /**
164
+ * Verify likely direct-image links when possible. Origins commonly block
165
+ * server-side preview agents, so a strong file extension remains a useful
166
+ * fallback: the client can still request the original resource itself.
167
+ */
168
+ private fetchDirectImage;
169
+ /** Fetch and validate YouTube's official preview metadata. */
170
+ private fetchYouTubeEmbed;
171
+ /**
172
+ * Last-resort thumbnail for a Facebook post whose page carried no
173
+ * Open Graph image: the media image rendered by Facebook's public embed
174
+ * plugin, which is served to third parties without a login wall.
175
+ */
176
+ private fetchFacebookEmbedImage;
177
+ private fetchTweet;
178
+ private fetchReddit;
179
+ /**
180
+ * Fetch Reddit's fixed-host JSON endpoint without the Undici dispatcher.
181
+ * The platform child runtime has exhibited stuck Undici requests even when
182
+ * AbortSignal fires. node:https gives us an independently enforced socket
183
+ * timeout; the outer withDeadline still guarantees callback completion.
184
+ */
185
+ private fetchRedditJson;
186
+ private fetchRedditEmbed;
187
+ private scrape;
133
188
  cleanup(cb: PlatformCallback): void;
134
189
  }