eurafaeldecarvalho/skills

A collection of Claude skills. Each skill lives in its own folder with a SKILL.md.
## Skills
Finding what to build — recurring agents that hunt and validate ideas before you build:
| Skill | What it does | Scope |
| apps-ideas-routine/ | A daily autonomous routine (Grok Bot, Claude scheduled tasks or any agent with a persistent workspace) that starts from who already pays and complains, keeps all state in SQLite with database-enforced guards, validates without talking to users (public data → fake-door landing → paid test → real checkout → paywalled MVP) and only then proposes building. Ships with schema, queries, a tested calc lib and 66 sourced benchmarks. | Web & mobile apps |
| games-ideas-routine/ | A daily autonomous routine for small games (web portals like Poki/CrazyGames, Steam, mobile, Roblox/UEFN) that starts from loops already proven (platform arbitrage, genre cohorts, jams) plus a structural twist, keeps all state in SQLite with database-enforced guards, validates by measured player behavior (portal playtests → Steam page/demo → paid tests only when they buy a real sample), picks the engine per idea and aims at a portfolio in net R$/hour, not MRR. Ships with schema, queries, a tested calc lib, 65 sourced benchmarks and a 352-test suite. | Games |
| oss-ideas-routine/ | A daily autonomous routine for open-source repos meant to earn GitHub stars, not money. It starts from measured demand (issues with hundreds of 👍 the maintainer refused, abandoned popular tools, tool requests on HN), cuts by category hit rate and maintenance load, keeps all state in SQLite with database-enforced guards, runs a launch-readiness gate, and judges each launch on organic stars after the owner's own posts (E14/E30/E90, two channels before a kill, net of any pre-launch pace). Ships with schema, queries, a tested calc lib, 63 sourced benchmarks and a 128-test suite. | Open source |
Once you're building — together these are the "impeccable + taste-skill, but for games":
| Skill | What it does | Scope |
| game-feel/ | How the action FEELS — the feedback/juice layer (easing, number-pop, trauma-based screenshake, hit-stop, particles, sound). | Shared (2D & 3D) |
| game-taste-2d/ | How a 2D game LOOKS — art direction (palette, value-based legibility, composition, mascot, camera/parallax, HUD/typography, anti-slop). The "taste-skill for 2D games". | 2D |
| game-taste-3d/ | How a stylized/low-poly 3D game LOOKS — art direction (palette, lighting & post, materials/shaders, silhouette & modeling, 3D camera, web/Steam pipeline, anti-slop). The "taste-skill for 3D games". | 3D |
| game-polish/ | The whole-build audit/polish pass (feel, art, readability, onboarding, performance, mobile, accessibility, audio, the shareable moment) → a prioritized fix list. The "impeccable for games". | Shared (2D & 3D) |
| roblox-game-qa/ | The Roblox QA + refinement pass to run before publishing/updating an experience or after a playtest. Starts with a mandatory shared-machine step (read the shared schedule, RAM pre-flight, book a slot, coordinate with other agents, Blocked when no slot), then ordered passes (audit, play critique, harden, red-team, adapt, polish, FINAL PASS on the published build, feedback loop) repeated in a convergence loop until two clean iterations. Covers build/tests (Rojo, Luau checks, Lune, a Lune fake server for multiplayer), every device and input (phone, tablet, desktop, console; touch, mouse/keyboard, gamepad), ProximityPrompts and camera, game flow, server authority and remotes, visuals/audio, measured performance, safety, publishing, accessibility and localization. Ships scripts (verify.sh, a Studio MCP play harness, a UI audit probe, an Output scanner, box-preflight.sh) and templates (context, feature inventory, report with iteration ledger, bug, manual checks). 0-4 scores, P0-P3 severity with examples, an anti-pattern catalog, a case study and a release gate with real-device and real-player checks. Method adapted from Impeccable. | Roblox |