Review an AI-generated spec with non-engineers

Have Claude draft the spec as an HTML doc — diagrams, tables, callouts and all. Paste the artifact at duodoc's home. You get a URL.

The rich HTML preserves Claude's formatting (mermaid diagrams, code-with-language-hints, warning callouts). The parallel Markdown view at ?view=review strips it to clean prose for reviewers who want to skim.

Slack the URL. Reviewers open it. If the visibility is link, they don't need an account to read. When they want to comment, they sign in with one magic-link email — no password, no Google OAuth dance.

The interaction model is the one they already know from Google Docs and Notion: select text, type, post. The card lands in the right gutter, aligned with the selection. No surprises.

Every comment is anchored to a fingerprint of the selected text (the surrounding context, not just an offset). When the doc changes — even structurally — the comment either re-locates itself or gets flagged as "anchor not found" rather than silently pointing at the wrong line. No more "wait, what was this comment about?" weeks later.

Replies thread one level deep. Comments resolve. Resolved threads stay in the gutter at 50% opacity so an auditor can see what was asked and how it was answered.

When a comment is concrete ("add a section on rate limiting," "the rollback step needs an explicit owner"), the doc owner can click Address with Claude on the card. Claude rewrites just the affected region with the comment as context, returns the proposed v2 in a sandboxed preview, and waits for accept-or-discard. The original commenter sees the updated version with their comment resolved.

Once the conversation winds down, download:

• .html for a polished artifact to attach to a PR.
• .md for the version that lands in Notion, Confluence, or wherever specs live in your org.
• .zip bundle with both files plus the open-comments sidecar — useful for an archived record of what was discussed before sign-off.