evolve-workspacelisted

# Evolve Workspace Turn explicit evidence into the smallest durable improvement. Do not infer preferences from silence, clicks, dwell time, retries, or other ambiguous behavior. ## Capture The Signal 1. Preserve the user's exact words or a faithful summary with `feedback.record-explicit`. Attach the narrowest known target: current Artifact, Case, scenario, Skill, or Workspace Profile. 2. Separate scope before changing anything: - “这次”“这一篇” is Case-specific unless the user says otherwise; - “以后”“我的文章都” is a lasting preference; - a factual correction fixes the responsible content or Skill regardless of preference; - an upstream release is only an update candidate, never implicit consent to replace local truth. 3. Do not manufacture learning signals from objective runtime data whose meaning is unclear. Failed commands may diagnose a tool problem, but they do not reveal a human preference. ## Locate Responsibility 1. Compare the feedback with the Goal, the stable Benchmark or temporary Case acceptance evidence, and the Artifact chain. 2. Identify the smallest owner: current Artifact, Profile, one business Skill, one scenario Benchmark, or one Workflow condition. Do not edit all of them to make the decision look important. 3. Read the complete owner Skill package before changing its method or `## 完成标准`. If the Skill has `SOURCE.md`, preserve upstream identity and record local changes. ## Change And Reprove 1. First repair the current Case at the responsible Arti