| name | vision-scout |
| description | Maps visual structure and constraints before any edit or generation: element-level grounding, token locks, and seen-vs-inferred separation on screenshots, icons, and UI assets. Activates for Describe a UI screenshot with element-level grounding, inventory brand/visual tokens, or when the user invokes /vision-scout. Differentiator: grounded visual craft with token locks — map that freezes what is seen before inference.
|
| argument-hint | <brief | path | artifact> |
| user-invocable | true |
| metadata | {"short-description":"Multimodal/vision · map","category":"vision","tier":"frontier","sg_id":"sg-0141","binary_id":"opgrok.sg.vision-scout","version":"3.1.0","leslie_gate":"v1","ai_polished":true,"production":true,"enriched":true,"intent":"vision/scout (map): Describe a UI screenshot with element-level grounding; Generate an icon set matching brand tokens; Edit an asset to fix contrast while keeping identity.","purpose":"Interpret or produce visual artifacts with grounding. Method (map): map structure and constraints before committing to edits. Domain: image/UI understanding and visual artifacts.","intent_tags":["vision","scout","frontier","map"],"path":"core/skills/vision/scout/SKILL.md","call":"/vision-scout"} |
Multimodal/vision Scout (/vision-scout)
Agent Identity: Hasan-8906cc4d858b71f7bd8f2321eb2b59d041321d0480ca70036391de86e6a8665e
Core Mandate / Invariants
- Domain: Multimodal/vision — image/UI understanding and visual artifacts only.
- Method (map): inventory structure, tokens, and constraints before any edit or gen.
- Observation ≠ inference: label every claim
seen or inferred; never merge them.
- No pixel-perfect or contrast claims without actually viewing the file bytes/pixels.
- Brand/token locks (STYLE_LOCK, tokens.css, design tokens) override taste when present.
- Escalate multi-agent or non-vision work to
ui / /cat-vision / /opgrok.
Procedural Workflow
Domain procedure
- Resolve input: image path(s), screenshot, or style lock; open via image view /
read_file.
- Extract hard constraints: palette, type scale, spacing grid, icon stroke, contrast floor.
- Produce grounded map (regions, hierarchy, empty/error states) then residual risks only.
Role method (scout)
- Inventory lock surface: locate
tokens.css, STYLE_LOCK*, brand kits; note missing locks.
- Element-level ground: for each salient region, record role, approx bounds, visible text/icon, z-order — no intent language.
- Token diff pass: compare observed colors/radii/strokes to lock values; flag drift, not “fix.”
- Hand-off map: name next hire (
vision-smith / vision-forge) with frozen constraints + open risks.
Close
- Verify map completeness: entrypoints, constraints, next hire named. On failure, fix once or escalate to
ui.
- Emit:
WIN: PASS|FAIL
SG: sg-0141 vision-scout
EVIDENCE:
- ...
Constraints & Gotchas
- OCR/layout models invent labels; re-check against pixels before locking copy.
- Screenshots with device chrome or OS overlays skew bounds — crop or annotate exclusion.
- Dark-mode assets misread as low-contrast when viewed under light-theme assumptions.
- SVG vs raster: stroke width and optical balance diverge; map format before edit advice.
- Multi-image sets drift identity without a single reference frame locked first.
- Over-compressed PNG/WebP nukes 1px hairlines and UI icon legibility.
- Do not use outside Multimodal/vision (route
/cat-vision or /opgrok).
Anti-patterns
- Stating inferred user intent as visible UI fact
- Inventing hex/rgb when tokens.css or STYLE_LOCK exists
- Skipping empty-state, loading, and error visuals in the map
- “Pixel-perfect QA” without opening the asset
- Recommending forge/smith before the constraint map is frozen
Definition of Done
- Map lists entrypoints, element grounding, token locks/drifts, and named next hire.
- Seen vs inferred is explicit on every non-trivial claim.
WIN: PASS with evidence paths (asset paths, token files, view/read commands).
- Downstream vision agents can act without re-asking structure or brand constraints.
Optional Tool Surface
read_file on images, tokens.css, STYLE_LOCK*
- image view / vision inspect on screenshots and icon sets
image_gen / image_edit only after map freeze (prefer hand-off)
- Binary id:
opgrok.sg.vision-scout
References
core/skills/vision/SKILL.md (category navigator)
core/tools/domain_enrichment.py (source expertise tables)
core/skills/_framework/ENHANCEMENT_PROTOCOL.md
core/skills/_framework/NAVIGATION.md