When you need depth, not a quick answer. Parallel agents search Reddit, X, Academic, Podcast, Substack, Web, and YouTube.
When you really need substance on a topic. Market analysis, competitor research, a new industry, pricing comparisons, discovery for a client, prep for a pitch. Not for quick lookups, that's what the personal assistant is for.
Research needs room. Its own session.
The skill starts a dialog. You answer a few questions: what exactly you want to know, how deep, which platforms.
Parallel agents go to work, each on a source. Takes a few minutes. At the end you get a synthesized report per agent plus a master synthesis.
Reports land in outputs/deep-research/{date}-{slug}/. You can revisit them anytime, forward them, or feed them into the next workflow.
Clean session end. Reports are safe in the workspace.
Community sentiment, honest opinions, what actually frustrates people.
Real-time discourse, thought leaders, current debates.
Peer-reviewed papers via OpenAlex. When you need substance.
Transcripts from relevant episodes. Deeper context.
Newsletters and long-form analysis.
Video transcripts, walkthroughs, explainers.
First a recon scan: where does the signal on this topic live? Then multiple Claude agents are dispatched in parallel, each on one source. Each agent goes three layers deep, scores signal, triangulates, and runs a critic review on its own findings. Finally: a master report that consolidates the individual reports.