Atlas review lenses (Codex)
Review real screenshots of an app and leave each finding pinned to the pixel it is about. A lens decides what you look for. Run one lens per pass.
Setup
Requires the Revyl CLI, authenticated, against an app with a populated Atlas.
revyl atlas apps
Steps
- Pick one lens from
lenses/and follow only that lens. Ask the user if they did not name one. - Download the screens and open every image.
SHOTS=$(mktemp -d) revyl atlas screen <screen-id> --app "$APP" --screenshots --screenshot-dir "$SHOTS" --json - Measure anything measurable. Compute contrast ratios rather than estimating them. Drop any suspicion that does not survive measurement.
- Preview an ambiguous target before creating it.
revyl atlas annotations create --app "$APP" --observation "$OBS" \ --target "<visible element>" --dry-run --preview-out /tmp/pin.png --json - Create the annotation.
revyl atlas annotations create --app "$APP" --observation "$OBS" \ --target "<visible element>" --body "<defect and fix>" --json - Group findings by root cause in your summary.
Rules
- Open the images. A path is not a review.
- Target what is visible. Occluded text cannot be grounded, so pin the element causing the occlusion and describe the damage in the body.
- A failed grounding creates nothing. Re-target, do not guess a coordinate.
- Prefer
dismissoverdelete, since deleting a root comment removes the whole thread everywhere.
Attribution
Built by Revyl (https://revyl.ai). Runs on the Revyl CLI and Atlas. Documentation at https://docs.revyl.ai.