Ctx research
Agent skills I use day to day (lowband / handoff / multi-agent-driven / ctx-* family)
npx -y skills add keli-wen/skills --skill ctx-researchAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
3 things to look at
- 20 days oldThe repository was created 20 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
基于项目持久 context 体系的深度调研。触发语：「ctx research」「基于 context 调研 X」「结合我们的 contexts 做深度研究」。先读已有 contexts，把问题磨到真正的知识缺口上，再多智能体展开调研，关键论断做对抗验证，结论写回去。
SKILL.md
2.0 KB, 593 tokens by cl100k_base, as published. Nobody here has run it
ctx-research
从「项目已经知道什么」出发的调研，以「项目知道得更多」收尾。它和通用 deep research 的差别有两处：搜索之前先用已有 contexts 把问题磨尖，验证过的发现要写回 context 体系。
语言：跟随用户的语言。
步骤
- 落地：找到项目的持久 context 层（比如带地图文件的
contexts/树，或 docs/ADR 体系）。读与问题相关的条目，把读到的分成三类：已回答的（不再调研）、假设过但没验证的、真正开放的。 - 磨题：把调研问题重述为「用户需要的」减去「contexts 已有的」的差值。只有差值本身模糊时才和用户确认范围，否则直接推进。
- 展开（用
multi-agent-driven的模式）：不同角度的并行调研者——官方文档、社区与先例、论文、真实代码。每人返回可引用的原始发现，带来源名和 URL 或路径。 - 验证：会驱动决策的论断都过一遍对抗检查：由独立的一步尝试反驳它或证明它过时。关键论断标注 confirmed / plausible / refuted。
- 综合：一份报告——先答案，再证据，最后开放问题。引用的事实和你自己的判断分开呈现。
- 写回：持久的部分交给
to-ctx，按项目规则放进 context 体系。聊天里给摘要加文件路径。
规则
- contexts 已回答的永不重复调研；引用那个 context 文件即可。
- 验证力度跟决策影响成正比，不跟论断有多有趣成正比。
- 依赖方向：可以用
multi-agent-driven、to-ctx和handoff；永不调用面向用户的 skill。
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.
Gives 0 of the 12 instructions most research analysis skills give in 593 tokens
Counted across 1,063 of the 1,754 authors here whose files we hold, read 2026-08-07
- Generate a markdown reportin 32 of 1063, across 23 files
- Cite each claim's sourcein 30 of 1063, across 15 files
- Define the ideal customer profilein 20 of 1063, across 2 files
- Search for companies matching the criteriain 20 of 1063, across 2 files
- Assign a fit score from one to tenin 20 of 1063, across 2 files
- Analyze the codebase to understand the productin 19 of 1063, across 1 file
- Ask clarifying questions about the value propositionin 19 of 1063, across 1 file
- Look for signals of immediate needin 19 of 1063, across 1 file
- Identify the target decision maker rolein 19 of 1063, across 1 file
- Suggest a personalized contact strategyin 19 of 1063, across 1 file
- Provide conversation starters for outreachin 19 of 1063, across 1 file
- Format results in a scannable markdown templatein 19 of 1063, across 1 file
Said here and by no other author read
- restate the question as a knowledge gap
- confirm scope only if the gap is ambiguous
- run parallel researchers from different angles
- return findings with source names and URLs
- run adversarial checks on decision-driving claims
- label key claims confirmed plausible or refuted
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.