task-researchlisted

# Task research You are reducing uncertainty *before* code is written or a plan is drafted. Research is cheap now and expensive later - an hour here saves a day of building the wrong thing. Timebox it and produce a brief, not an essay. ## Rule 0 - search the codebase before the web Most "how do I build X" questions are already answered inside this repo. - Grep for existing utilities, patterns, and near-duplicates of what you're about to add. Reuse beats new code. - Find the *convention*: how does this codebase already do the adjacent thing? Match it. - Cite what you find as `path:line`. A claim without a location is a guess. - Only reach for external sources once you've confirmed the answer isn't already here. ## External research - **Triage sources by trust:** official docs and the library's own repo/issues > maintained examples > recent blog posts > forum answers. A 2019 Stack Overflow answer about a fast-moving library is probably wrong now. - **Check version and date.** Pin every claim to the version it applies to. APIs drift; "the top Google result" is often two majors behind. - **Corroborate load-bearing claims.** If a decision hinges on it, find a second independent source. For a deep, fully-cited investigation, hand off to the `deep-research` skill instead. ## Separate facts from assumptions - Label every statement: **fact** (verified, with a source/location) vs **assumption** (plausible, unverified). - Make unknowns explicit. "I don't know whether the API sup