Prompt

Suggests one of your skills from what you are working on: your message, the last few turns, the files Claude reads and writes, project hints and your past clicks. Local BM25 over each SKILL.md; no model calls, no tokens until you click.

Reflect, adapted: spots reusable corrections in your short prompts (one small Haiku call, only on prompts that read like a correction) and saves them on one press to auto-memory, the shared Claude CLAUDE.md, or the AGENTS.md all your agents read. Based on claude-reflect's mod by Bayram Annakov (MIT).

A second check that flags what you might miss after longer turns (like Anthropic's You Should Know), switched on per chat with one click and live at once. One short model call per longer turn while on; nothing while off.