← Back to the Marketplace

Daily picks · November 22, 2026

Warm Personas, Careful Helpers, and a Safety Net

This week's standout bots are opinionated about their craft and careful about your data: a warm butler that commits to one answer, a persona engineer, an Obsidian librarian, and a new safety gatekeeper.

The Butler With a Backbone

大肥鱼 is a warm, human-flavored chief assistant that stays gentle without slipping into customer-service tones. It says what it means, keeps answers to one to three sentences, leads with conclusions, and refuses to hedge by offering both A and B — it gives one recommendation. Three hard lines hold it together: no fabrication, no bluffing, no performance. For anyone tired of chatty assistants that cannot commit, it is the most-installed bot this week.

大肥鱼

The Obsidian Librarian

Built on the LLM-Wiki method, this bot organizes and maintains an Obsidian knowledge base. It imports sources while preserving the originals, writes summaries, and links related topics, concepts, and entities. Ask a question and it checks the base and sources first, cites supporting evidence, and discloses uncertainty. A review mode flags broken links, orphan pages, duplicates, missing sources, conflicts, and stale content. It backs up before changing anything and never silently overwrites conflicting sources. For anyone with a sprawling vault, it is the most thoughtful cleanup crew around.

Obsidian 知识库治理

Fix Your Bot's Personality

If your bot still sounds like a press release, this persona engineer is the final answer. It ships a 29KB handbook covering a three-iron-rule gate, a five-part diagnostic before any rewrite, and a long list of techniques: translate vague asks like be livelier into observable behavior, convert attributes into mechanisms, and treat output rhythm as character. Before delivery it runs a checklist and a dialogue simulation so you know the card works. It reads as methodology you can reuse for any character, not just its own.

Tobs人格助手

The New Safety Gatekeeper

New this week, 拦截卫生 is a security gatekeeper for DSH. It asks three questions before anything moves: who initiated it, should they have this permission, and can the change be undone. It judges dangerous operations, reviews approval requests, and maintains the interception rule library, but it never executes risky actions or pretends to have blocked what it could not. Its real enforcement points are the cc-safety-net plugin, approval cards, and the file sandbox; the bot supplies judgment, evidence, and rule upkeep. Any request to remove a plugin, edit the registry, or loosen the sandbox needs your explicit nod.

拦截卫生

Past picks