Kenny Ponders肯玩 AI
← Projects
02 / ThinkResearch skillClaude Code + Codex

weave

Find the frame that makes a question clear.

When a smooth summary hides where the ground is weak, weave keeps the evidence visible and lets the material choose its shape.

Read the repository ↗证据先行,取景框决定文章
Frame Decisionthe article is the last step
xmaterial
Eevidence
f₁ · f₂ · f₃candidate frames
f*selected frame
f*(x)article

A frame earns its place when it changes what the article keeps, how it moves, and where it stops.

A readable answer can still hide its ground.

Research agents are good at making scattered material sound coherent. The dangerous part is what disappears in the smoothing: disagreement, weak evidence, and the boundary where a claim stops working.

weave treats the article as a view over evidence, not as a summary of everything the model can say.

The material changes the lens.

One pipeline, three ways into a question. The route decides what counts as evidence before any frame is allowed to lead.

01

deep-read

Papers, articles, interviews, reports

Evidence-grounded Chinese longform

02

source-dive

Repositories, frameworks, technical projects

Behavior paths, state owners, failure boundaries

03

survey

Domains and research directions

A map whose structure follows the source set

A frame is useful only when it earns the crop.

Candidate count is not the goal. A frame survives only when it is faithful, specific, selective, honest about its limits, and meaningfully different from the alternatives.

  1. 01Evidence fidelity

    Every load-bearing part can be traced back to the evidence model.

  2. 02Specificity

    It cannot be moved unchanged onto the next source or neighboring topic.

  3. 03Selection effect

    It changes the evidence, chapter order, causal story, or predicted boundary.

  4. 04Boundary honesty

    It names the condition, counterexample, or evidence gap that limits it.

  5. 05Distinctness

    It is not a synonym wearing a different title.

The honest part is part of the product.

weave does not promise a perfect view. It makes the crop inspectable, then leaves the edge visible.

It does
  • Collect sources before composing.
  • Reserve evidence and test the selected frame against it.
  • Separate verified behavior from static inference.
It does not
  • Turn one thin source into a field-wide claim.
  • Keep a frame alive just to manufacture a choice.
  • Replace the archived Loom thinking and vault utilities.

Bring a question with a little weight.

Install the skill, give it your material, and see which frame the evidence can actually carry.

想找点什么?

StudioProjectsNotesAboutSupport 支持创作Macao Route Atlas Explore Macao through an interactive heritage route.Fun Hacker News A daily Chinese digest of AI and agent discussions.Script-Weaver Turn a story premise into scripts and storyboards.card-skill Turn content into visual work, with quality checks built in.VideoWipe Remove burned-in subtitles, watermarks, and overlays from video.Kami Reader Companion A continuous reading and editing workspace in Obsidian.Kami Reader Bring the feel of reading on paper to Obsidian.Tracework Keep the reasons behind the changes.Clipplane Keep the words that matter on your own computer first.image-scrub Local image cleanup and screenshot simulation.social-content-kit Turn sources into social copy and visual briefs.Git Daily Note Updater Build daily notes from Git and agent sessions.weave Find a framework that makes sense of the question.loom-skills The archived predecessor of weave.BuddyBar A small companion in your Claude Code statusline.storyboard-prompter Use clear constraints for coherent characters and action.ask-back Choose useful questions for a reverse interview.Taku Give coding agents sound engineering habits.Human System Optimization An extended guide to human system optimization.GraphRAGxplorer Explore graph-based retrieval through visualization.Open-OmniSearch A retrieval agent that plans as evidence arrives.Sohu Text Matching 2021 A competition solution for Chinese text matching.torchKbert A customized PyTorch BERT library for experiments.Agent 搜索系统的盲区Codex 为什么要和 ChatGPT 融合:实现成本下降之后,知识工作的中心变成了什么Anthropic 提出的 J-space 真正证明了什么?伴读 · AI智能体上下文工程:Manus构建实践与启示伴读:Perplexity 团队如何设计 Agent Skill?伴读:你能读懂的强化学习Anthropic 提出的 J-space 又是个什么东西?Agent Memory Systems:用四个控制面看懂长期记忆文件系统即新数据库:我为 AI 智能体构建个人操作系统之路【译·读】Hermes Agent 深度解析ChatGPT 的 Dreaming记忆做了什么Beyond the Prompt:Claude Code 深度指南Claude Code: Session, Context 和 Memory 管理机制深度解析Claude Code Compact 机制研究:从上下文压缩到可恢复代理记忆为什么同样的模型在 Hermes 上比在 OpenClaw 上更聪明?Claude Code 完全拆解:从 52 万行源码看 AI 编程助手的架构设计