SL: read_gyazo verb — fetch a shared image from a Gyazo link

[JDHayesBC]

From the deck, 2026-08-25. Gyazo is the de-facto way images are shared in SL.

Idea

A read_gyazo(url) verb in sl.py: take a Gyazo URL → strip to the ID → fetch https://i.gyazo.com/{id}.jpg (or .png) directly → save locally and return the path so it can be viewed.

Gotcha (proven live 2026-08-25)

The https://gyazo.com/{id} page URL is HTML — feeding it to image handling throws 400 "Could not process image" (both Lyra and Caia hit this). You MUST hit the i.gyazo.com/{id}.<ext> direct image. Verb should do the strip+direct-fetch and try .jpg then .png.

Acceptance

• read_gyazo("https://gyazo.com/<id>") → downloads the raw image, returns a local path.
• Handles both page-URL and already-direct forms; falls back .jpg→.png.
• Foundation for the auto-prefetch issue (perception intercepts links).

[JDHayesBC]

Shipped in sl.py (2026-08-26, terminal-Lyra).

read_gyazo(url) is a staticmethod (pure HTTP — no Corrade/creds needed, so it works even when logged out). It:

• strips any Gyazo URL shape to the id (gyazo.com/<id>, i.gyazo.com/<id>.png, bare id, trailing slash / query string)
• fetches the direct image i.gyazo.com/<id>.<ext>, trying jpg → png → gif (Gyazo doesn't say which up front)
• guards on Content-Type: image/* so an HTML error page (the 400 "could not process image" gotcha) is skipped, not saved
• saves to haven/data/gyazo/<id>.<ext> and returns {success, path, url, id, bytes}

Tested offline: id-extraction across all URL shapes ✓; CLI end-to-end on a bogus id → clean {success:False, error:'.gif: Not Found'} with no crash ✓; missing-arg → usage+exit 2 ✓.

One live check remains before closing: a REAL shared gyazo URL → downloads a viewable image (logged in IN-WORLD-TESTING.md). Foundation for the #302 perception auto-prefetch is now in place.