Suggest mode for your AI drafts.
Turn a markdown file into a self-contained browser review surface: select text to leave Suggest / Comment / Delete annotations anchored as Tufte-style margin sidenotes, then export structured markdown for an AI to apply.
Reviewing long text with an AI shouldn't feel like dictating edits over the phone.
You ask an AI for a draft. It writes a good one. Now you want to push back: tighten this sentence, cut that section, ask a question in the margin. So the copy-paste dance starts. Read in one window, type "in paragraph three, change X to Y" in another, lose your place, forget half of what you meant to flag.
annotate makes the page itself the review surface. You read and mark up in one place. The notes come back as structured markdown your AI can act on without you re-explaining a thing.
- Inline marks, margin notes. Suggest an edit, leave a comment, or strike something out, all anchored to text position.
- A clean handoff. A tidy handoff doc for your AI to make the exact changes you want.
- Zero infrastructure. The output is one HTML file. Open it anywhere, no server.
- Built to read in. warm off-white paper, a book serif, and a quiet margin inspired by Edward Tufte.
annotate is a standard SKILL.md skill, so it works in any agent.
Paste this to your agent and let it install itself:
Install the annotate skill from https://github.com/jonnilundy/annotate-skill. Read the repo and set yourself up to use it the way that's native to your environment, then use it whenever I ask to review or annotate a markdown file.
Or use the universal installer (Claude Code, Cursor, Codex, and ~50 more agents):
npx skills add jonnilundy/annotate-skillWorks with pnpx too. It detects your agent and drops the skill where that agent looks for it.
Or add it as a Claude Code plugin:
/plugin marketplace add jonnilundy/annotate-skill
/plugin install annotate@annotate-skill
Once it's installed, ask your agent to make a markdown file reviewable:
"annotate
{file}.md""make this reviewable"
"create an annotation page for this doc"
- The skill writes
{file}.review.htmlnext to your original md file and opens it. - Select text → choose between Suggest · Comment · Delete.
- Each note appears as a margin sidenote tied to a numbered mark.
- Add high-level direction in Notes for Claude at the top.
- Click Copy Changes (or Export for a file) and give it back to your AI.
- Your AI applies the edits and answers the comments.
| Type | Meaning | Looks like |
|---|---|---|
| Suggest | replace the selected text | blue underline with the new text |
| Comment | a note or a question | yellow-gold highlight |
| Delete | remove the selected text | red strike-through |
The whole tool is one HTML file. No build and nothing to keep running. It honors your original markdown exactly: the review file is generated straight from your .md, and the export maps back to it line by line. And because the handoff is a compact list of just the changes, not the whole document, it stays token-efficient when you give it back to your AI.
Most review tools are loud. This one is quiet on purpose. The text sits in a narrow, readable column and your edits live in the margin beside it, the way notes have lived on paper for centuries. You can scan a whole document and see every mark in context without a panel covering the words you're reviewing. Borrowed, gratefully, from Tufte CSS.
- Python 3 to generate the review file
MIT licensed. See the MIT License.