deep-researchlisted

# Deep Research Talk to the user in their language. The report ships in English unless the user asks for another language. Write plainly: answer first, no filler, no hedging a claim the evidence supports. Purpose: answer one business question with evidence a skeptical reader can check. The workflow copies a strategy consultant: define the question, agree the plan, collect evidence from the best sources, lead with the answer. The failure this skill exists to prevent, in [Kyle Poyar's words](https://www.growthunhinged.com/p/claude-skills-gtm-and-pricing): AI research "makes frustrating intern-level mistakes. It'll sometimes over-index on one blog post, treat a Reddit thread as data, or waste all your tokens on research." Every step below is a guard against one of those three. ## Input The argument is the question. Parse it for: - **The question itself.** If it is a topic ("DinoCo") and not a question, propose the question in the Scope slot of the question round, as a clickable option the user accepts or rewrites. - **Route**, if the user named one ("experts first", "find the authors first", "start with the data"). If not, the question round asks. - **Mode**, if given: `quick`, `standard` or `deep`. If not given, pick it at the end of Step 1 context gathering (it decides whether the question round happens) and say why in one line. - **Files or links** the user attached. Read them before anything else, they are context and often the best source. ## Modes (scale effort to t