adopt
Take a folder that already holds research — papers, notes, code, half-written drafts — into MAGI, without breaking what is already there.
Talk a run over with the person until it can be written down — motivation, directions, what is not worth it, when to stop, their taste — and they sign it.
$ npx -y skills add Misaka16384/Wikify --skill discuss --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/discussContext preview
The summary Claude sees to decide when to auto-load this skill.
Talk a run over with the person until it can be written down — motivation, directions, what is not worth it, when to stop, their taste — and they sign it.
name: discuss description: "Talk a run over with the person until it can be written down — motivation, directions, what is not worth it, when to stop, their taste — and they sign it." commands: discuss: "Align with the person and write the contract of a run." origin: magi
`magi next` opens with `DISCUSSING`, or the person wants something explored while they are away. Nothing is authorised in this phase: the job is to talk.
1. `magi run start --title "<what it is after>"` opens the run and its empty contract in `threads/`. Read what the project already knows first (`ask`). 2. Converge on a few directions. Each becomes a `[[question]]`; a guess the person holds becomes its own conjectured proposition carrying their bet: `magi thread bet <slug> <supported|refuted|unknown>`. 3. Fill the contract's sections in the person's language and, where you can, their words: Motivation, Directions, Not worth it, Stop when, Taste, Allowed. 4. End on an alignment quiz — 3 to 5 hypothetical forks about risk: "only a known result in new words: go on or drop it?", "the worker's safe first choice or its radical last one?". The answers go under Taste, verbatim. 5. Read the contract back to them. They sign, or say so and you type it: `magi run sign <run> --steps N --mode <deep|balanced|explore>` — the number and the mode are theirs to say (deep: a chain that stands; explore: a map).
here: work done before the signature is in no trajectory and nobody authorised it. Reading, ingesting and compiling seed papers is fine.
not written down is lost the moment another agent takes the run over.
contract is all it gets. Write it for that reader.
`magi run amend <run> --text "<what changes>"`, edit, sign again.
MAGI 是一个 agent-native 的科研工作环境:人是驾驶员,LLM agent 是机体,确定性的 magi CLI 是拘束具——三者同步率越高,科研越快。它把学术论文(PDF/LaTeX)摄入、编译为 Obsidian 兼容的概念卡片项目,并用三核架构管理完整的科研状态: 进入任意项目先跑 magi sync——它输出同步率、三核状态和逐条可执行的修复提示。magi radar 是文献雷达:定时发现相关新论文,并侦察"知识上应引用我方论文却未引用"的候选。同一份 skills 通吃 Claude
Take a folder that already holds research — papers, notes, code, half-written drafts — into MAGI, without breaking what is already there.
Answer a question from this project's own knowledge — retrieved, read and cited, never from memory.
Write the one thing a person reads after an unattended run — its report — and the lecture notes that prepare them for it, in words they already have.
Compile raw sources into dense, interlinked reference and concept cards, and mine the concepts a compiled card left implicit.
Turn whatever the human has — a PDF, a link, a DOI, a screenshot of a citation — into raw/ sources this project can compile.