What problem does it solve? Teams lose track of which decisions were made, which still need human judgment, and which are just automatic follow-ups. This Skill audits an existing Loopbench project and produces an evidence-backed decision report before asking anyone to decide anything. ## Core Features & Use Cases - Deterministic decision scan: Runs loopbench decisions scan to collect accepted contracts, recorded human gates, and process gates as facts, with optional bounded local Codex/Claude session reading via --with-sessions. - Lead classification: Classifies Git surface signals and needs-analysis items into current_human, automatic_followup, or past_inferred using explicit observable criteria, with mandatory evidence references and tradeoff options. - Validated report rendering: Writes analysis.json, validates it against the scan (rejecting bad references, duplicates, or transcript leakage), and renders report.html separating explicit past decisions from inferred ones. - Use Case: A maintainer returning to a long-running project asks "what decisions did this project make and what still needs my judgment?" The Skill scans the repository, separates direction-setting choices from automatic follow-ups, and returns an HTML report with cited evidence. ## Quick Start Ask the assistant to scan this Loopbench project for past decisions and open questions that still need human judgment.
