skilld
All skills
thedivergentai avatar

/godot-agent-vision

@90634df
by Divergent AIthedivergentai/gd-agentic-skills783 stars
48

Agentic eyes for Godot work: cross-platform screenshot CLI (asset/region/window/screen) to budgeted WebP, TEMP editor viewport bridge under .gdskills/, and Taste Receptor Atlas v2 (~237 micro-receptors × 0–2: hierarchy, contrast, typography×41, affordance, space, color, icon, composition, light/FX, motion-tells, anti-slop, originality, identity). Use when agents must see UI, assets, or the editor — not for in-game screenshot Autoloads. Keywords: agent vision, screenshot, WebP, visual QA, taste receptors, typography, UI review, asset sheet, editor viewport, anti-slop.

Ask your Agent

Use this Skill: https://skilld.dev/gh/thedivergentai/gd-agentic-skills/godot-agent-vision

This session only. Nothing lands on disk.

referenceswebp-budgets.md

≈440 tokens on demand. Your agent reads this file only when SKILL.md points to it.

WebP budgets for agent vision

Encode captures as lossy WebP before the model Reads them. Tokens track resolution; request size tracks codec — PNG walls waste latency.

Defaults (CLI) — conserve context

Knob Value When
Short (lowest) edge 512 px (--short-edge) Default — identity, element overview, layout-at-a-glance, anti-slop tells
Quality 80 (--quality) Default encode
Method 4 (Pillow) Default encode
Soft size re-encode if > 1 MB Safety net
Contact sheet ≤ 2×2 tiles, ~12 px gutter Still short-edge fitted after compose

Never upscale small assets.

When to raise budget (detail / full layout)

Only when 512 short-edge is not enough:

Need Flag
Zone in on glyphs, inspector text, 1px seams --detail (long edge 1568) or --max-edge 1568
Dense full-screen layout audit --detail or --max-edge 1568–2048
Still blurry after detail tighter --region crop first; then --quality 90 
# overview / identity (default)
python scripts/capture.py window --title Godot

# details / OCR / full-layout
python scripts/capture.py window --title Godot --detail
python scripts/capture.py region --x 100 --y 100 --w 800 --h 600 --max-edge 1568

Agent rules

  • Prefer 1–3 WebPs per review turn
  • Start at short-edge 512; escalate to --detail only after a fail that needs pixels
  • Prefer one sheet over many single icon files
  • Prefer window/region over 4K/5K full desktop
  • After Read + rubric, prune old timestamps if cluttered (keep last few)

Implemented in webp_encode.py and asset_sheet.py.

Source: SKILL.md on GitHub

Third-party checks

1 warning1mo3 checks · Risk MEDIUM
  • Gen Agent Trust Hub1mo

    The skill captures screenshots and interfaces with the Godot editor by executing system commands and modifying project configuration files. It includes a temporary editor bridge staged as a Godot plugin, which introduces risks related to command execution and dynamic module loading.

  • Socket1mo

    No alerts

  • Snyk1mo

    Risk: LOW · No issues

Provenance

Signed by skilld at 90634df. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 3 weeks ago.

Activeupdated 2 months ago

README badge

README badge for thedivergentai/gd-agentic-skills/godot-agent-vision