feed-watch

feed-watch: collect → score → drafts → digest

Everything specific to this project — which feed, what counts as an item, the rubric, the threshold, the labels, the reply template, the copy gate — is in .set/feed-watch.yaml and the files it names. Read it first: set-feed-watch show prints it resolved.

Iron rules

• Read only. You never post, comment, react, like, share, join, follow, message or send anything, on any source. A human sends every reply.
• Only the declared feeds. Open only the url of a declared source and the permalinks of its items. No messenger, notifications, profiles, search, or other groups or pages.
• Human pace, one tab. Scroll a screen at a time, wait 2–4 seconds, read what is visible. No parallel requests, no scraper, no JavaScript bulk extraction.
• On trouble, stop and report. No refresh, no retry loop, no workaround for a login wall or a challenge.

0. Preparation

1. Ask the user to bring the Chrome window to the front (logged in to the source) and keep it there during the run. Measured: in a background window the feed rendered one item and then only loading placeholders.
2. set-feed-watch show — sources, caps (max_items_per_run, max_age_days), the item filter file, the rubric guidance file, labels, draft template and blocks.
3. set-feed-watch seen — the ids and text hashes already collected.
4. Load the Chrome tools in ONE ToolSearch call (see the claude-in-chrome skill), open one new tab on the source's url. Apply the source's sort_hint; if you cannot, say so in the report.

1. Collection

Scroll in small steps. Stop when ANY of these holds, and record which one:

• you reached the newest item already collected, or items older than max_age_days;
• you processed max_items_per_run items;
• stall: two consecutive scrolls brought no new item — only loading placeholders;
• challenge: a login wall, captcha, verification prompt, or an error page.

A run that collected zero items while the feed showed only placeholders is a stall, never "no new items".