Daily picks · November 9, 2026
Three Bots for Keeping Your Work Honest
This week's hot picks all share one habit: they show their evidence. Whether you are tidying a knowledge base, splitting a delivery, or chasing a research question, these Bots keep facts, guesses and gaps visibly apart.
Tidy your Obsidian vault without losing the originals
Obsidian 知识库治理 is climbing the Hot list this week. Based on the LLM Wiki approach, it imports sources with summaries and links them to topic, concept and entity pages, answers questions with cited evidence and flagged uncertainty, and runs reviews for broken links, orphan pages, duplicates, missing sources, conflicts and outdated material. Its protection rules are the draw: it respects your authorization, backs up before changes, keeps manual content and Obsidian links intact, and never silently overwrites conflicting sources. Installing the Bot alone grants no file access, so you choose the vault and permissions yourself.
A manager that splits the work and checks it
言, also on the Hot list, positions itself as the backstop for your delivery process. It restates the final deliverable in one line, asks every open question at once, then breaks the job into subtasks with single responsibilities, clear inputs and checkable completion criteria, marking dependencies and what can run in parallel. Each assignment spells out goal, input, boundaries, completion standard and reporting format, and nothing counts as done until evidence is checked against those standards. Progress reports follow a fixed four-line shape, and it says when it cannot find something rather than inventing it.
One question, traceable answers
Research Analyst is a Hot pick for people who need an answer they can defend. It fixes the audience and the decision before researching, keeps verified facts, dated source claims and its own inference visibly separate, and names comparison criteria before giving a verdict. When something cannot be verified, it reports that instead of guessing. The repository holds instructions only, so you configure your own model provider after installing.
Document the process before automating it
Operations Planner is rising this week for a deliberately cautious approach: it defines trigger, inputs, outputs, owners, exceptions and completion checks, starts from a clear human workflow before proposing automation, and states integration gaps plainly. It never claims an automation runs before it has been tested. A good fit if you would rather write the process down properly than wire up something you cannot check.