tsilva

Generates professional logos with programmatic transparency conversion. Works with pixel art, vector designs, and complex multi-colored styles. Use when asked to "generate a logo", "create logo", or "make a project logo".

Create, modify, validate, and optimize README.md files following GitHub best practices.

Analyzes a codebase and generates a pure requirements specification (SPEC.md). Use when asked to "extract requirements", "create a spec", "generate a blueprint", or "distill this project".

Applies gorgeous terminal styling to bash scripts using gum with ANSI fallback. Use when creating or modifying bash/shell scripts that produce user-facing terminal output. Use when asked to "style bash output", "make pretty", "add colors to shell script", or "improve terminal UX".

Applies gorgeous terminal styling to Python scripts using Rich with plain-text fallback. Use when creating or modifying Python scripts that produce user-facing terminal output. Use when asked to "style python output", "make python pretty", "add colors to python script", or "improve python terminal UX".

Tune one checked-in rlab SB3 PPO or A2C recipe from durable training completion signals without launching checkpoint evaluations. Use when the user points to a recipe and asks to tune, optimize, autoresearch, improve sample efficiency, maximize training return, find the best hyperparameters, or make training behavior stable across seeds. Runs a bounded fixed-rung beast-3 search, confirms the winner on five untouched training seeds, and patches only the pointed leaf recipe.

Launch, continuously monitor, and when explicitly authorized repair queue-backed rlab training from a checked-in goal and recipe, defaulting to beast-3. Use when the user asks to launch, run, start, execute, follow, watch, monitor, diagnose, fix, or harden an rlab training recipe, run, or research goal. Stay attached through terminal events, report W&B URLs immediately, preserve failed-run evidence, and return final statistics, the best artifact, and the exact playback command.

Start and execute a persistent repair-mode goal that stress-tests the complete rlab queue, fleet, training, checkpoint-evaluation, W&B publication, artifact, playback, and cleanup lifecycle. Use when the user explicitly asks to harden, stress-test, repeatedly exercise, or fix the fleet experiment lifecycle until it is reliable, especially with bounded SuperMarioBros-NES PPO runs across Level1-1 through Level1-4.

Composite rlab release workflow for trained checkpoints. Use when the user asks to upload, publish, release, or promote a checkpoint/model to Hugging Face, especially RL checkpoints with gameplay preview videos. Coordinates Hugging Face model-card publishing with video preview requirements and YouTube upload of the same preview video.

Upload or update YouTube model preview videos from rlab, especially RL gameplay previews associated with Hugging Face model cards. Use when the user asks to upload a local video to YouTube, update an existing video description/title/tags/privacy, add a video to a playlist, produce OAuth authorization flow instructions, or ensure model-card preview videos have clean clickable model links.

Operate a hosted game in the scummweb repo to achieve an in-game objective using only Codex's native browser control, screenshot reasoning, local references, and verified hover labels. Use when the user asks to play, progress, solve a puzzle, test a gameplay path, or complete a game without OpenRouter, external vision APIs, image generation, or image segmentation.

Use when working in the scummweb repo and the user asks you to achieve an in-game objective, progress a puzzle, test a gameplay path, or operate a hosted game through browser automation. This skill covers a minimal-action gameplay workflow, early walkthrough lookup, control discovery, save-state reuse, and game-specific references such as Beneath a Steel Sky inventory and verb handling.

Use when working in the scummweb repo and the user asks to scrape, add, release, deploy, or verify another freeware game from the official ScummVM site. This skill covers sourcing the official download, enabling the matching engine, wiring launcher art and metadata, uploading the game payload to R2, refreshing the managed bundle, verifying launch, and deploying to production unless preview is requested.

Use when working in the scummweb repo and the user asks to add, replace, refresh, or verify game artwork, cover art, posters, hero images, screenshots, or dashboard assets for one or more games. This skill covers sourcing appropriate online art, storing it locally, wiring it into the dashboard metadata, preserving the repo's managed ScummVM bundle workflow, and verifying the result with Playwright.

Generate a dated what-next report for a selected live healthpilot profile. Use when the user wants a prescriptive "what should I do next?" answer, a current action report, or an updated report after new labs, exams, visit feedback, or symptom changes. The report should include both unresolved-issue actions and health-optimization actions when the data supports them.

Generate a dated Markdown report of active medications, past medications, and supplements from a selected live healthpilot profile. Use when the user asks for a medication list, supplement list, medication history, or a written medication/supplement report in .output/.

Ask a selected live healthpilot profile's 10 highest-yield unanswered questions interactively, then write a paste-ready Markdown health-log entry under .output/ for the user's real health log.

Generate a dated root-cause differential report for a selected live healthpilot profile. Use when the user asks for the most likely causes of a symptom, condition, abnormal marker, episode, pattern, or health problem, especially when they want top-K hypotheses with explicit probabilities that add to 100% and explanations for each probability.

Launch and monitor a SuperMarioBros-Nes-turbo PyPI release. Use when the user says /build-release, asks to cut a release, asks to tag/publish a version, asks whether a release made it to PyPI, or asks for supermariobrosnes-turbo release artifacts.

Regenerate the verified Stable Retro versus SuperMarioBros-Nes-turbo Level 1-1 promotional video and README GIF preview. Use when asked to rebuild, refresh, rerun, or update the Mario throughput comparison, its exact shared GymRec trajectory, local benchmark numbers, verification evidence, final tracked MP4, or animated GIF preview.

Lightweight Super Mario Bros NES throughput optimization loop for this repo. Use when optimizing, profiling, benchmarking, or iterating on emulator speed around env_steps_per_sec without feature or accuracy regressions, including repeated +10% cumulative improvement rounds resumed from AUTORESEARCH_ROOT_PATH state.

Use before any Waxon fix or behavior-changing edit, especially behavior regressions, product-flow issues, auth/deploy/UI bugs, or changes where user-stated specs or the overarching product objective could be lost. Reads SPECS.md and keeps the objective, failure, causal hypothesis, non-goals, and proof path aligned before editing.

Run a Waxon browser-use test suite from a markdown file. Use when the user asks to run, rerun, execute, validate, or fix a Browser Use or signed-in user experience suite and provides or references a `.md` test-suite file such as `tests/browser-use-signed-in-user-experience.md`.

Use when making Waxon design, styling, layout, responsive UI, visual polish, or frontend interaction changes. Loads the repo-local design-system reference and approved UI comparison workflow before editing the Next.js app.

Generate, fix, audit, and wire avatar assets for the aipit project. Use when Codex needs to source or process still preset avatars, crop or convert portrait images, generate speaking avatar videos, repair broken avatarUrl or speakingAvatarUrl paths, update avatar attribution, validate avatar media, or audit built-in avatar coverage.

Add or update debate character presets for the aipit project. Use when Codex needs to create a new participant character, extend the Portuguese or global roster, add or revise character relationships, add search/test coverage for preset discovery, or wire a new preset into starter bundles.

Create, optimize, fix, or audit an individual AI Grounds learning playground and its AI Guide. Use for new or planned concepts, live playground URLs or slugs, beginner-learning audits, deterministic lesson fixes, and full teaching-experience optimization. Keep cross-lesson ordering, prerequisite-lesson creation, and curriculum splits or merges in the separate lesson-plan review skill.

Use when reviewing, optimizing, or reconciling the AI Grounds dashboard lesson plan, including checking live-vs-planned status, adding implemented lessons missing from the dashboard, reordering lessons into the smoothest learning sequence, adding missing prerequisite lessons, and breaking oversized lessons into smaller playground-sized steps.

Use when reviewing Parselabs extracted lab rows for a profile. By default, review and accept/reject only unconfirmed rows through the Browser Use review UI or helper artifacts. Also supports explicit full re-audit of accepted/rejected rows, rejected-row root-cause investigation, source-evidence comparison against page/bbox crops, patching extracted page JSON when values or review decisions are proven wrong, and rebuilding profile outputs.

Code quality and clarity guidelines for Python - enforces flat orchestrator patterns, explicit error handling, and mandatory comments

Audit a ModelArchViz model by opening the local app with the official OpenAI Browser Use plugin, clicking architecture nodes one by one, and verifying that each selected architecture part highlights the correct source code lines in the code pane. Use when checking model architecture specs, code highlight mappings, expanded/lazy architecture groups, generated model-code artifacts, PyTorch/JAX language parity, or regressions in the ModelArchViz UI.

Author/edit/review/normalize neural-network model files/snippets: Python, PyTorch, JAX/Flax, app/model-notebooks sources, generated/displayed model-code, readable tensor-shape-aware teaching code.

Comprehensive quality assessment for document extraction pipelines. Use when (1) After implementing extraction pipeline, (2) Before production deployment, (3) Periodic quality audits, (4) Investigating suspected quality issues, (5) Validating pipeline changes, (6) Benchmarking extraction accuracy across document types/eras. Performs stratified sampling, systematic verification, pattern analysis, and generates detailed quality reports.

Comprehensive quality assessment for document extraction pipelines. Use when (1) After implementing extraction pipeline, (2) Before production deployment, (3) Periodic quality audits, (4) Investigating suspected quality issues, (5) Validating pipeline changes, (6) Benchmarking extraction accuracy across document types/eras. Performs stratified sampling, systematic verification, pattern analysis, and generates detailed quality reports.

Use when verifying claudesk after changes to ensure core functionality works end-to-end before committing.

Use when TypeScript errors may have been introduced, after editing .ts files, or when debugging unexpected runtime behavior that could be a type mismatch.