The Draftmark skill wraps the dm CLI so your coding agent knows the whole loop — publish a plan, share the link, read the feedback, iterate, close the review — without you spelling out a single API call. Install it once; it activates on the right prompts.
Clone the repo and drop the draftmark/ folder into ~/.claude/skills/ (or a project's .claude/skills/). Restart your agent — the skill is discovered automatically. No per-project setup.
Ask your agent to "share this plan for review." The skill fires, checks auth with dm whoami, writes the markdown to a file, and runs dm create --agent — then hands you the share link.
Reviewers open the link and leave comments — general or anchored to specific lines — in the browser. Nothing to install on their end. The agent just waits; review is asynchronous.
Next session, ask "any feedback yet?" The skill resolves the doc from .draftmark.json, pulls comments as JSON, incorporates them with dm update, and closes the review when it's done.
Activates on the prompts that matter — "share this for review," "get feedback on this draft," "any comments yet?" — and stays out of the way otherwise.
No new runtime — the skill drives the same dm CLI you can run by hand. Auth, JSON, and exit codes are already solved.
Reads results with --json and parses structured output — slugs, line-anchored comments, review status — instead of scraping tables.
Knows the exit codes: 2 auth, 3 not found, 4 conflict (review closed or past deadline). It reports gates instead of retrying blindly.
.draftmark.json is the memory between sessions. A fresh session picks up the pending doc and its feedback without you re-pasting a link.
When your agent reviews someone else's doc, it comments with --author-type agent and line anchors, so its feedback carries a visible [agent] badge.
Checks dm whoami before acting and asks for an account key when one is missing — no silent failures, no leaked keys in tracked files.
Honors --base-url, so the same skill works against a self-hosted Draftmark instance without any changes.