THE AGENT SKILLS DIRECTORY
Reusable skills for AI agents. Install any of them into a Cohub world's agent with one command.
npx -y @neta-art/neta-search-cli@latest skills find "<what you need>"
Wrote a skill? Publish it here →
Basic Cohub abilities. Most agents already have these mounted; listed here for reference.
Paste this into your agent's chat — it fetches the skill and follows it.
Use the skill: production-tile-collision-map. Run `npx -y @neta-art/neta-search-cli@latest skills use neta-skills/production-tile-collision-map --via site` and follow what it prints. Read its complete output, redirecting it to a temporary file first if necessary, and resolve relative paths from the skill directory it prints.npx -y @neta-art/neta-search-cli@latest skills use neta-skills/production-tile-collision-mapTurn your map art into playable collision grids with roads, open areas, and interactive path testing.
# Production Tile Collision Map Skill ## Hard default For new production map backgrounds, **use image generation by default** with `references/map-generation-prompt.md`. Do **not** create SVG diagrams, abstract block layouts, or placeholder vector art as the shipped `background.png`. SVG/mock diagrams are allowed only as internal explanatory sketches when the user explicitly asks for a schematic. Always preserve/use the collision review template in `templates/collision-preview/` for public QA review. ## Current default pipeline: latest recognition-led differential clearance ```text image-model background.png -> background QA -> image-recognition-candidates.json # primary semantic source: image-derived road/open/blocker candidates -> raw visual candidates # bounded connector/blocker override hints only -> recognition-source-audit.json # proves source shift and clean regeneration -> road-grid.json # image-filtered road + critical connectors -> open-area-grid.json # image-filtered open area + blocker keepout -> walkable-union-grid.json -> road-clearance-grid.json # road: 3-tile clearance -> open-area-clearance-grid.json # open area: 5-tile clearance -> union-clearance-grid.json # 5-tile semantic-union baseline -> clearance-grid.json # runtime = unionClearance5 ∪ roadClearance3 -> walkable-grid.json # compatibility alias -> route-cost-grid.json -> turn-aware A* path QA -> semantic-diff-report.json # compares recognition source vs overrides / prior candidates -> latest-recognition-contract.json # declares source priority and forbidden shortcuts -> 06-final.tmj -> HTML collision preview ``` Core rule: do not claim a latest rebuild if only QA/report files changed. A latest rebuild must regenerate `image-recognition-candidates.json` and every downstream grid/export from the current background. Runtime convention: ```text 0 = walkable 1 = blocked ``` ## Required workflow ### 1. Generate or accept production background - For new maps, call an image-generation model; do not hand-build SVG/vector placeholder backgrounds. - Use `references/map-generation-prompt.md`. - Prefer `1024x1024` or another clean tile multiple. - Require top-down orthographic camera, grid-readable roads, clear water/roof/road/open-area boundaries, no UI/text/labels/characters. - If the image fails QA, regenerate/edit first rather than forcing collision. ### 2. Semantic layers Split walkability into: - `road-grid`: stone roads, dirt lanes, bridge decks, door approaches, narrow branch paths. - `open-area-grid`: courtyards, plazas, broad grass/ground planes. - `blocked`: roofs/buildings, water, walls, dense trees, fences, props. Latest source-priority contract: 1. Treat `image-recognition-candidates.json` as the primary semantic source. 2. Use manual/raw visual shapes only as bounded hints for critical connectors, thresholds, bridges, doors, and hard-blocker overrides. 3. Filter manual road/open hints against image-recognized candidates; do not let broad rectangles become the actual semantic layer. 4. Merge image-recognized hard blockers with explicit blocker overrides before clearance. 5. Output `recognition-source-audit.json` and `semantic-diff-report.json` so reviewers can verify that semantic recognition was actually rerun. Road detection should emphasize continuity and route structure, not only color. Open-area detection should emphasize area and openness, not narrow strips. ### 3. Differentiated clearance Default: ```text roadAgentDiameterTiles = 3 openAreaAgentDiameterTiles = 5 runtime = unionClearance5 ∪ roadClearance3 ``` This keeps the safe broad-body baseline while additively preserving narrow roads. When iterating, compare with the previous runtime grid: ```text newWalkable > 0 newlyBlocked == 0 # for additive road-clearance improvements ``` ### 4. Route cost and pathfinding Use `route-cost-grid.json` and turn-aware A\*: ```text stepCost = 10 turnCost ≈ 80 roadCost < openAreaCost nearObstaclePenalty > 0 blockedCost = 9999 ``` Report raw tile length, compressed segment count, turns, reachability, and layer-specific QA probes. ### 5. HTML collision preview Use `templates/collision-preview/` as the default review page. It includes: - canvas map preview; - click-to-path; - WASD / arrow movement; - grid toggle; - collision toggle; - QA point toggle; - semantic layer toggle `L`; - turn-aware route HUD. Publish this preview to public share after each meaningful revision. ## Expected output package ```text background.png annotations.json image-recognition-candidates.json recognition-source-audit.json raw-walkable-grid.json road-grid.json open-area-grid.json walkable-union-grid.json road-clearance-grid.json open-area-clearance-grid.json union-clearance-grid.json clearance-grid.json walkable-grid.json route-cost-grid.json semantic-diff-report.json latest-recognition-contract.json 06-final.tmj qa/path-qa.json qa/collision-overlay-probe-qa.png index.html style.css script.js ``` ## References - `references/map-generation-prompt.md` - `references/semantic-layered-clearance.md` - `references/runtime-integration.md` - `references/annotation-schema.md` - `templates/collision-preview/`