Delegates tasks to Codex, Claude, Pi — or, on demand, Cursor, Antigravity (agy), and Grok — via the yoyo CLI. Use for multi-agent coordination, best-of-n comparison, independent review, deep research, fresh-context loops, scheduled/continuous work, and agent-to-agent handoffs.
Generate real raster images (diagrams, explainer illustrations, plan visuals, mockups, posters) through yoyo imagegen, which renders with GPT-image (gpt-image-2). Use when a document, plan, HTML report, or explanation would land better with a visual, or when the user asks for an image, diagram, illustration, or mockup.
Reasoning-discipline harness for delegated work — contract before acting, evidence for every claim, and a done gate before any completion claim. Ships inbuilt with yoyo and is injected into every non-raw delegation by default; set YOYO_DEFAULT_SKILLS to override it or to the empty string to disable.
Deep research on a question or decision via yoyo research — parallel cross-vendor lenses (proponent/skeptic/analyst/explorer/pragmatist or your own) synthesized into a decision brief that surfaces disagreement. Use for "should we X?", technology choices, roadmap questions, and any decision worth multiple independent perspectives.
Save and rerun multi-agent Yoyo pipelines from a JSON spec — parallel fan-out phases, cross-checks, and implement-then-review chains. Use when an orchestration has stabilized and is worth rerunning by name; for one-off coordination, orchestrate directly with yoyo ask/research/review.
Understand any video (URL or local file) — summarize it, answer questions about it, diagnose a bug from a screen recording, break down someone's content. Runs the bundled watch pipeline (yt-dlp + ffmpeg + captions/Whisper) directly, or delegates the whole watch to a yoyo agent so frames never enter your context. Use when a task involves a video: YouTube/Loom/TikTok/X links, .mp4/.mov/.mkv/.webm files, screen recordings.