Benchmark or evaluate an entire AI model. Create, list, inspect, and compare WorldJen Bench runs — large-scale evaluation across many prompts and dimensions, executed on a GPU worker queue. Invoke ONLY when the target is a whole model — a model id, a checkpoint, a Hugging Face repo, "the model", "my model", comparing two models, tracking drift across model versions, or gating CI on a model's benchmark. To score an individual clip, video, or image, use `worldjen-score` instead, NOT this skill. Backed by `worldjen bench create / list / get / cancel / delete / logs / csv / videos / download-videos`. NOT for runner host setup (use `worldjen-runner`) or ranking clips that share a prompt (use `worldjen-rank`).
Install or upgrade the WorldJen SDK and CLI in the active Python environment. Use when setting up `worldjen` for the first time, switching between core and runner installs, verifying the install, or upgrading to the latest `worldjen` / `worldjen[runner]` release (auto-restarts the runner systemd service so it picks up the new version). NOT for runner host registration (see `worldjen-runner`), benchmark run lifecycle (see `worldjen-bench`), or upgrading the worldjen-skills plugin itself (see `worldjen-update`).
Fetch the public WorldJen leaderboard. Use when the user wants model rankings, available evaluation dimensions, or a comparison of top entries. No auth required.
Rank or compare several generated clips that share one prompt into a leaderboard. Upload candidate videos or images to the user-scoped Rank session and get them sorted best-to-worst; the session locks to the first prompt (server-enforced). Invoke ONLY when the user explicitly asks to rank, sort, compare, or pick the best among multiple clips, variants, or candidates for the same prompt. Backed by `worldjen rank get/upload/current/reset`. For raw scores on a single clip use `worldjen-score`; for benchmarking an entire AI model use `worldjen-bench`.
Set up and operate the WorldJen runner host — the GPU worker daemon that pulls evaluation jobs and executes them. Use when registering, starting, stopping, or inspecting a Linux+systemd machine that runs WorldJen jobs. NOT for evaluation run lifecycle (use `worldjen-bench` for that).
Score an individual generated clip, video, or image. Upload one (or a few) media files to the user-scoped Score session and get raw per-dimension quality scores back, with optional custom prompt and dimension selection. This is the DEFAULT for "score it / score this / score these" whenever the target is a piece of media — a clip, video, image, shot, frame, generation, or output — and NOT an entire AI model. Use for quick feedback on a generation, A/B-ing two clips, or inspecting evaluator scores without a full benchmark. If it is unclear whether "it" means a single clip or a whole model, ask the user before scoring. Backed by `worldjen score get/reset/upload`. For ranking several clips that share a prompt use `worldjen-rank`; for benchmarking an entire model use `worldjen-bench`.
Upgrade the worldjen-skills plugin to the latest release. Use when an `UPGRADE_AVAILABLE` banner appears in another skill's preamble, or when the user asks to update / upgrade the worldjen skills.