Протокол автономного AI-исследования

★ 7.5 · research

autonomous-investigation is a Claude Code skill that defines the canonical protocol for autonomous AI research: a hard question budget (typically 3 clarifying questions), a mandatory search-plan gate before any research begins, three-level evidence labeling (Fact / Inference / Assumption) on every key claim, a do-not-invent list of domain-specific fabrication risks, and a stable output schema that keeps consecutive runs diffable. The skill requires no input by itself — it establishes the contract that other investigation skills follow; when invoked standalone, providing a research target and the decision it should support is enough to proceed. Each run closes with exactly four numbered next-step options. It is built for competitive scans, scheduled intelligence refreshes, and any research task where distinguishing sourced facts from inference matters and unattended execution is required.