BotDoc Translator

The official skill for turning what you've already built — a project, a chat transcript, or a folder of notes — into a properly-structured BotDoc your agent can build from.

Hand this to your LLM along with whatever you want to publish, and you'll get back a BotDoc draft you can paste straight into the editor at botdocs.ai/new.

When to use this

• You vibe-coded an app and want to share the spec so others can build their own version.
• You have a chat transcript with Claude or ChatGPT where you worked something out, and want to capture what you learned.
• You have a folder of notes, READMEs, or scattered markdown describing a system you've put together.
• You have an idea fully fleshed out in your head and want it structured before publishing.

What's in this BotDoc

File	What it does
index.md	This file — install paths and quick start.
system-prompt.md	The raw prompt. Paste it cold into any LLM.
examples.md	Three before/after pairs so the LLM has shots in context, and you can see what good output looks like.
claude-skill.md	Drop-in Claude Skill (Anthropic format). Rename to SKILL.md and place under ~/.claude/skills/botdoc-translator/.
claude-code-command.md	Drop-in Claude Code slash command. Rename to botdocify.md and place under ~/.claude/commands/. Triggers on /botdocify.
cursor-rule.mdc	Drop-in Cursor Rule. Place under .cursor/rules/ in any project.
chatgpt-gpt.md	Configuration recipe for publishing as a Custom GPT in the ChatGPT Store. Paste field-by-field into the GPT Builder.

How to use it

Claude Skill (Claude.ai / Claude Code / Claude Desktop)

The fastest path for Claude users.

1. Open claude-skill.md from this BotDoc.
2. Save it as SKILL.md inside ~/.claude/skills/botdoc-translator/ (or a project-scoped <project>/.claude/skills/botdoc-translator/).
3. Restart your Claude session.
4. Ask Claude "Compile this project into a BotDoc" — the skill activates automatically.
5. Paste the JSON it returns into botdocs.ai/new → Import from JSON.

Claude Code slash command

If you live in the terminal.

1. Open claude-code-command.md.
2. Save it as botdocify.md inside ~/.claude/commands/ (user-wide) or <project>/.claude/commands/ (project-scoped).
3. Run /botdocify from any Claude Code session — Claude reads your codebase directly, no copy-paste.

Cursor / Windsurf / Zed

Cursor Rule format works in Cursor; the same file pasted into Windsurf or Zed's project rules works there too.

1. Open cursor-rule.mdc.
2. Save it under .cursor/rules/botdoc-translator.mdc at the root of any project (or your editor's equivalent rules directory).
3. Ask Cursor "Use the botdoc-translator rule to compile this project into a BotDoc."

ChatGPT (Custom GPT)

The cleanest path for ChatGPT users — once it's published, anyone with a ChatGPT account can use the GPT directly from the GPT Store, no setup.

If a public BotDoc Translator GPT already exists, just open it and start chatting. If you want to publish your own copy (or maintain a private fork):

1. Open chatgpt-gpt.md from this BotDoc.
2. Follow the step-by-step setup — paste each field into the GPT Builder (Name, Description, Instructions, Conversation starters), upload examples.md as a knowledge file, and toggle the Web Browsing + Code Interpreter capabilities on.
3. Publish, share the URL.

Generic (any LLM — Gemini, Llama, anything else)

1. Copy the contents of system-prompt.md.
2. Paste it as the system prompt in your chat (or as the first message labeled "System:").
3. Send your raw material — your README, codebase summary, chat transcript, or notes. More context in = better BotDoc out.
4. The LLM will reply with a JSON object.
5. Open botdocs.ai/new, click Import from JSON, and paste.

CLI (advanced)

botdocs clone @botdocs/botdoc-translator
cat system-prompt.md | claude -p "$(cat your-project/README.md)" \
  | botdocs publish -

Quality tips

• More context wins. Paste your README, your top-level files, and a brief description of what you built. For chat transcripts, paste the whole thing — the LLM will compress it.
• Tell the LLM the audience if it's not obvious from the input. "This is for backend developers who want to self-host a Notion clone."
• Always review. LLMs fabricate when underfed. Check the description, tags, and license before publishing.

Limitations

• LLMs are non-deterministic. Two runs on the same input can produce slightly different drafts. That's fine — you're going to edit either way.
• Smaller / older models drift from the schema. Use Claude Sonnet 4.6+, GPT-4 class, or Gemini 2.5 Pro for best results.
• Your data stays in your LLM session — this BotDoc never calls Botdocs servers. The result is yours to paste, edit, or discard.

Contributing

Found a case where the prompt fails? Open a Prompt Request on this BotDoc with the input that broke it and what you'd have expected — that's how the prompt improves.