Design safe autonomous research/experiment loops inspired by Karpathy's autoresearch. Use for bounded experiment harnesses, optimization loops, overnight research plans, benchmark exploration, or repeated hypothesis testing with explicit budget, rollback, logging, and approval gates.
Plan and produce concise product brag/demo videos, launch clips, or social cutdowns from an app, feature, workflow, or release. Use when the user wants a shareable product video, feature teaser, before/after demo, founder update, or launch asset. Inspired by latent-spaces/brag, but rewritten for Overdrive with no raw upstream redistribution.
Safely guide authorized website rebuilds and visual recreations from a URL, including when to use JCodesMore's ai-website-cloner-template as a fresh project template. Use for "clone this website", "rebuild this landing page", "make a site like this", or "replicate this design" requests.
Check factual claims in writing against credible sources, rate support levels, and suggest safer wording or better citations. Use for blog posts, landing pages, reports, social posts, research summaries, investor materials, SEO content, or any output with claims, stats, studies, or factual assertions.
Use when the user wants to extract or approximate a design language from a public or authorized website: colors, typography, spacing, components, Tailwind/shadcn tokens, visual patterns, and implementation guidance. Checks optional tooling first and falls back to manual browser/source inspection.
Use when the user asks about installing, updating, checking, or using Overdrive from Claude Code. This plugin is a thin helper; the full cross-agent skill system is installed by the Overdrive CLI.
Use for ambiguous, multi-step, or high-impact requests where the agent must clarify assumptions, compare 2-3 viable approaches, break work into phases, and get phase-one alignment before implementation. Trigger on vague "build", "refactor", "add feature", "figure out", product/design/dev asks, conflicting requirements, or work whose direction materially changes the outcome. Do not use for tiny factual answers or obvious one-line edits.
Use when the user references a local PDF, Office document, spreadsheet, presentation, HTML file, CSV/data file, or ovd-workflow knowledge-vault document and the agent should convert it to clean Markdown before reading to reduce tokens and preserve structure. Uses Microsoft's MarkItDown when available; falls back gracefully when Python/MarkItDown is unavailable.