Imported from noah-rivard/news-coverage-a24-lionsgate (extensions/chrome-intake/AGENTS.md). Install upstream with npx skills add noah-rivard/news-coverage-a24-lionsgate --skill chrome-intake. Copyright stays with the author.

Component Guide: extensions/chrome-intake/

Scope: Chrome Manifest V3 extension that scrapes article pages and posts payloads to the ingest API.

Gotchas and expectations:

• Build with npm run build (esbuild) to emit dist/. Load dist/ as the unpacked extension in Chrome. The build script now uses fileURLToPath so Windows paths (double drive letters) no longer break npm run build.
• Service worker stores the latest scraped article in chrome.storage.local and auto-sends the selected (latest) article to the configured endpoint in chrome.storage.sync (default http://localhost:8000/process/articles). The popup is mainly for status or manual resend.
• When using /process/articles, the popup now displays openai_response_ids returned by the server (when present) and includes quick links to open the configured endpoint, server /review and /health, and OpenAI Responses logs.
• Quarter is derived from the article date (prefers published_at, falls back to scrapedAt, then the current date) before posting; nothing is hard-coded in the payload.
• Content script uses @mozilla/readability; if a page blocks script injection, reloading may be required.
• Manifest requires only Feedly origins at install and lists all other origins as optional host permissions. The content script is no longer auto-injected; scraping happens on demand after a per-origin permission grant.
• Manifest must include the tabs permission so the service worker can open/close background tabs during link capture; without it tabs.create/remove calls fail.
• A single context menu item ("Capture article for ingest") is registered on page/frame/link. On click, the service worker requests host permission for the actual capture target (link destination or the iframe origin when right-clicking inside a frame) before injecting contentScript.js (background tab for links to avoid stealing focus). A 45s timeout closes background tabs that hang and reports a capture error (instead of failing silently).
• Update README/CHANGELOG when changing build steps, manifest, or permissions.
• published_at must be a YYYY-MM-DD date per coverage_schema.json; the content script trims any datetime meta tag (e.g., article:published_time) down to the date before sending to the service worker. If no publish date is found, the service worker now falls back to the scrape timestamp’s date before sending so both ingest and process endpoints have a valid date.
• If the endpoint contains /ingest/, the service worker sends the coverage-schema payload including a required facts array (min 1). Otherwise it sends the raw article (title/source/url/content/published_at) to the pipeline endpoint. When the endpoint contains /process/articles, it sends a single-item JSON array for the selected article (it does not batch previously captured items).