redline
Redline — the human→agent return path
Overview
Every other review path in this ecosystem runs agent→agent: a second model reads the work and judges it. This one runs the other direction. The human is the judge, and the bottleneck was never their judgment — it's the channel. Describing a wording change in chat costs more keystrokes than making it, and "the third paragraph feels off" throws away the exact phrase that caused the feeling.
/redline renders the artifact as a self-contained page where the human edits text directly and attaches comments to exact quotes, then returns one Markdown file with everything anchored.
No dependencies. One generated HTML file, everything inline, no network requests, no server, no npm or pip install, nothing to vet. It is the html skill's artifact rules plus a review layer.
If the current repo has its own .claude/skills/redline/ or .agents/skills/redline/, that version is authoritative — follow it instead. This global version is the fallback.
Announce at start: "Opening this for markup — edit and comment in the browser, then hit Send."
The rule underneath every step: an edit the human already made is a decision, not a suggestion. Carry their wording verbatim.