document-iteration-skill

Overview

Feedback given in chat gets lost. "Revise the part about the database" means nothing a week later, in a new session or on another device. With this skill, you write feedback directly in the document as %% markers %%. Claude edits the content and replies next to your comment. Everything stays in plain markdown, so git keeps the whole history.

Uses ==SQLite(DB)== for storage.

%%(DB) SQLite for v1 %%
•%%> Switched to SQLite: embedded, no server to run. <%%•

What it does

• Inline comments: %% comment %% asks for a change, %% ?: question %% asks a question, %% NOTE: context %% gives context, and %% APPROVED %% locks a section
• Precise targets: mark text with ==PostgreSQL(DB)==, then refer to it by name: %%(DB) switch to SQLite %%
• Inline replies: Claude answers with •%%> done <%%•, so the comment, the reply and the updated text stay together
• Plain requests: "respond to the comments in plan.md", "review this spec", "draft a proposal", "clean up plan.md". There are no commands to learn.
• A clean finish: a small Python script removes every marker and keeps the content. Its --check mode also works as a pre-commit or CI guard.
• Editor highlighting: ask Claude to set it up for VSCode or Obsidian, and your comments and Claude's show in different colors

Install

In Claude Code, copy the document-iteration-skill/ folder into .claude/skills/ in your project, or into ~/.claude/skills/ to use it everywhere. In Claude.ai, download the zip and upload it under Settings → Skills.

Install steps, the full syntax and editor setup are on the docs site.

Built with

Markdown, a SKILL.md skill definition, and one Python script for cleanup.