What problem does it solve? Research answers often sound confident but cannot be traced — the reader cannot tell which sentence came from a source, which came from the user, and which was invented. This Skill forces every important claim to carry an explicit evidence label and picks the lightest research path that actually answers the question. ## Core Features & Use Cases - Ask classification and path selection: Distinguishes local questions (answer in the repo), quick factual lookups, comparison memos, and deep-dive landscapes, then applies the matching depth instead of over-researching. - Evidence-labeled reports: Every claim is tagged [SOURCED], [USER], [INFERENCE], or [RECOMMENDATION], with dated citations, contrarian evidence, and an explicit confidence-and-gaps section. - Untrusted-source handling: Treats fetched pages as data, never instructions — flags prompt-injection attempts, single-source claims, and stale information. - Use Case: Ask "compare PostgreSQL vs MySQL for our workload" and receive a structured decision memo with 5-15 cited sources, a downside case, one recommendation, and a list of what could not be verified. ## Quick Start Ask the assistant to research a question such as "what is the current state of edge runtime cold starts" and request a cited report with evidence labels.
