Use an unprimed sub-agent as a second set of eyes before accepting visual work.\nThis is for visual defects, not pixel metrics; pair it with\ncompare-screenshots when you also need numbers — including on a single shot\nwith nothing to compare against, whose scene metrics say whether the frame has\nany content in it at all.
fork_context: false; pass only the full\nimages, the crops, and a short neutral task. Include the approved reference\nand user requirements when judging fidelity; withhold history, implementation\ndetails, prior verdicts and the expected answer.scripts/receipts.mjs. Every number it\nflags is a claim the critique cannot show a command for, and every file it\nflags is one the critique named but never wrote; strike those claims or\nre-measure them yourself before any of them reach a decision.Use this shape, replacing the bracketed surface and attaching local images:
\nFresh visual critique task. You have no project backstory and should only\ninspect the supplied screenshots and crops. First inspect the full screenshot\nfor context, then inspect each crop at zoomed scale. Look for concrete\nvisual/layout defects in [surface], especially unit/prop depth ordering,\nlayering, shadows, selection-marker contrast, ground-plane perspective,\nflag/pole attachment, label style/icons, blur, scale, lighting, artifacts,\nmissing models, terrain feature readability, roads, water, and scan\nreadability. Do not assume these are correct. Return a concise list of issues\nyou can see, with confidence and whether the issue is visible in the full image,\nthe crop, or both.\n\nDescribe what you see in words. You are looking at these images, not measuring\nthem, so state no pixel value, luminance, percentage, ratio, count, or offset\nunless you ran a command that produced it and you quote that command and its\noutput beside the number. \"The hem reads clearly darker than the floor beside\nit\" is a finding; \"44.0 versus 33.6\" without a receipt is a fabrication. The\nsame goes for files: name no artifact you did not create with a tool you\nactually have.\nSpawn config:
\nagent_type: explorerfork_context: falselocal_image itemsA number in a critique is a receipt or a guess, and the two are\nindistinguishable once they are written down. An agent handed images and asked\nfor defects will answer in whatever register the question was posed in: ask it\nhow much darker the hem is and it returns a luminance mean, because that is\nwhat the answer is shaped like — not because it measured anything. Sometimes\nthe estimate lands close and sometimes it inverts the finding; nothing in the\nreport tells you which, and it reads as the strongest evidence there precisely\nbecause it is numeric. The same reflex names the crops and dumps it says it\nwrote.
\nscripts/receipts.mjs\nover the returned report — node scripts/receipts.mjs <report.md> — before\nyou read the findings, so an unsupported number never gets to be persuasive\nfirst. It exits nonzero and prints each unbacked quantity and each file the\nreport named but never wrote. scripts/receipts.eval.mjs is its own eval,\nover reports written in the eval whose right answer is known by construction,\nboth directions pinned: run it after changing the checker, and never widen\nwhat counts as a receipt to make a failing case pass.compare-screenshots, or re-run the critic with\na shell and this same receipt rule. Never repair a report by asking its author\nto double-check its own figures.