LLM-inferred CLAUDE.md for new topics

[rsdrahat]

Summary

When a new Telegram forum topic auto-registers, use a lightweight LLM call to generate a tailored CLAUDE.md based on the topic name and first message content.

Current behavior

New topics get a smart template with:

• Topic name (from Telegram API)
• Obsidian sync instructions
• Keyword-matched container skills

Proposed enhancement

On first message in a new topic, make a single Claude Haiku call with:

• Topic name
• First message content
• List of available container skills + descriptions
• Parent group context

Generate a CLAUDE.md that:

• Describes the topic's purpose in 1-2 sentences
• References the most relevant skills with specific instructions
• Suggests an appropriate Obsidian folder for outputs
• Sets tone/behavior hints based on the topic's nature

Considerations

• Should be non-blocking (don't delay first response)
• Write a basic template immediately, then overwrite with LLM version async
• Use Haiku to keep cost minimal
• Only trigger on genuine new topics, not re-registrations

🤖 Generated with Claude Code