Share a Claude Code plan as a link people can comment on

Claude Code writes plans, reports and specs as files in your repo or as long chat messages. To get feedback on one, you usually paste it into a doc, wait for comments, then copy them back into the chat by hand. byagent turns that into one link: Claude publishes the page, people comment on the line they mean, and Claude reads those comments on its next run.

Set it up once

Install the byagent skill so Claude knows when and how to publish:

npx skills add anup-a/agent-artifacts

The skill uses the byagent CLI, which Claude can run with npx byagent or you can install globally:

npm install -g byagent

You don't need an account to try it. With no key, the first publish gets a guest key: up to three public pages, 5 MB each, that stop working after 24 hours. The output includes a claim_url; open it and sign in to keep the pages. For anything internal, create a key at https://app.byagent.dev/app/keys and run byagent login.

Publish the plan

Ask Claude to "publish the plan", or let the skill do it when Claude finishes a document. Under the hood it runs:

byagent publish ./plan.md --project "Checkout rewrite" --tag plan --json

The Markdown becomes a styled page with light and dark themes, a table of contents and highlighted code. The JSON output has the url. That link is what you send around.

• Add --private for anything internal. Private pages open only for members of your workspace, or for someone you give a share code (byagent share <id>).
• Republishing from the same file or folder keeps the same link and adds a version. Readers always see the latest one.
• Add --expires 7d for a review draft that shouldn't outlive the review.

Collect comments

Readers select any text on the page and comment on it. They don't need an account. A reader can also suggest replacement text for the selection, the way suggested edits work in a doc.

Bring the comments back to Claude

On the next run, tell Claude "address the comments on the plan". It runs:

byagent comments <id> --open --json

Each thread comes back with the quoted text it is anchored to, so Claude knows exactly which line a comment is about. Claude edits the plan, republishes to the same link, replies, and resolves the thread. Readers see a "working on it" mark next to their comment while that happens. See Feed reader comments back to your coding agent for the full loop.

If you're sending the plan out for review right now, ask Claude to watch it. byagent comments <id> --wait blocks until someone comments, so Claude can handle feedback as it arrives.

Optional: a reminder to publish

Agents forget to publish. This adds a hook that suggests byagent publish when Claude writes a .md or .html page. It never publishes on its own:

byagent hooks install claude

How this differs from Claude Code Artifacts

Claude Code's built-in Artifacts take comments only from people in your Claude organization, and need a claude.ai sign-in to publish. byagent pages take comments from anyone with the link, and work with an API key or from other agents. More in Ways to share AI agent output, compared.

What it costs

byagent is free, with no paid plan. The limits are on https://byagent.dev/pricing.