--image <png> --expect "<scene>" evaluates one shot; --shots runs every review shot in Screenshots/README.md.
Reads OPENROUTER_API_KEY from the ignored .env (never committed).
Returns {verdict, confidence, reason, missing}.
Documented in AGENTS.md (Visual review pre-check step), Tools/README.md, and MEMORY.md. It is a helper only — a human still approves.
Live run on main's older pre-warm-skin review-01-character.png correctly flagged it as missing the expected human figure (verdict "no", confidence 0.9) — a regression this is meant to catch.
Adds `Tools/judge_screenshot.py`, a small optional pre-check that uses OpenRouter multimodal image understanding (docs: https://openrouter.ai/docs/guides/overview/multimodal/image-understanding) to judge whether a screenshot shows the expected scene.
- `--image <png> --expect "<scene>"` evaluates one shot; `--shots` runs every review shot in `Screenshots/README.md`.
- Reads `OPENROUTER_API_KEY` from the ignored `.env` (never committed).
- Returns `{verdict, confidence, reason, missing}`.
Documented in `AGENTS.md` (Visual review pre-check step), `Tools/README.md`, and `MEMORY.md`. It is a helper only — a human still approves.
Live run on `main`'s older pre-warm-skin `review-01-character.png` correctly flagged it as missing the expected human figure (verdict "no", confidence 0.9) — a regression this is meant to catch.
- Tools/judge_screenshot.py: judges a screenshot against an expected-scene
description via OpenRouter multimodal image understanding, reading
OPENROUTER_API_KEY from the ignored .env. --image+--expect for one shot,
--shots to run every review shot in Screenshots/README.md. Returns a
{verdict, confidence, reason, missing} JSON.
- Documented in AGENTS.md (Visual review pre-check step) and Tools/README.md;
recorded the OpenRouter payload format gotcha in MEMORY.md.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Adds
Tools/judge_screenshot.py, a small optional pre-check that uses OpenRouter multimodal image understanding (docs: https://openrouter.ai/docs/guides/overview/multimodal/image-understanding) to judge whether a screenshot shows the expected scene.--image <png> --expect "<scene>"evaluates one shot;--shotsruns every review shot inScreenshots/README.md.OPENROUTER_API_KEYfrom the ignored.env(never committed).{verdict, confidence, reason, missing}.Documented in
AGENTS.md(Visual review pre-check step),Tools/README.md, andMEMORY.md. It is a helper only — a human still approves.Live run on
main's older pre-warm-skinreview-01-character.pngcorrectly flagged it as missing the expected human figure (verdict "no", confidence 0.9) — a regression this is meant to catch.- Tools/judge_screenshot.py: judges a screenshot against an expected-scene description via OpenRouter multimodal image understanding, reading OPENROUTER_API_KEY from the ignored .env. --image+--expect for one shot, --shots to run every review shot in Screenshots/README.md. Returns a {verdict, confidence, reason, missing} JSON. - Documented in AGENTS.md (Visual review pre-check step) and Tools/README.md; recorded the OpenRouter payload format gotcha in MEMORY.md.