Communitygithub.com

doublesq97-ui/know-me

AI 协作说明|让任何 AI 更懂得应该如何与你协作 · A portable AI Collaboration Manual

know-me とは?

know-me is a Claude Code agent skill that aI 协作说明|让任何 AI 更懂得应该如何与你协作 · A portable AI Collaboration Manual.

対応~Claude Code~Codex CLI~Cursor
npx skills add doublesq97-ui/know-me

お気に入りのAIに質問する

このエージェントスキルを事前に読み込んだ状態で新しいチャットを開きます。

ドキュメント

Know Me

Create a portable collaboration interface for the user: not a personality test, diagnosis, digital clone, or biography.

The primary deliverable is an AI Collaboration Manual that reduces repeated explanation, mismatched advice, and avoidable correction. Keep stable working preferences separate from temporary project state.

Route the entry

Choose exactly one branch from the user's wording. Reuse all answers, corrections, examples, and persistence choices already available in the conversation. Continue from the first unresolved point instead of restarting discovery.

First use

Use when the user invokes Know Me without choosing a path, or says Know Me 初次使用.

  1. Read first-use.md.
  2. Follow it completely. Inside an active Know Me run, interpret 开始 or 如何开始 as 快速开始.

Completion criterion: the user has chosen a branch. Do not begin collecting information before the choice.

Quick start

Use when the user chooses A, says 快速开始, or says 开始 inside an active Know Me run.

Read quick-start.md completely and follow it.

Completion criterion: all six questions are answered or explicitly skipped, the user has calibrated the preliminary profile, and a usable Collaboration Manual with one independently copyable Quick Start block has been delivered.

Deep calibration

Use when the user chooses B, explicitly asks for deeper exploration, or accepts an invitation after quick start.

Read deep-calibration.md completely and follow it.

Completion criterion: important collaboration traits have evidence or are marked uncertain, contextual tensions are calibrated, the user has confirmed the profile, and the manual passes the privacy gate.

Update an existing manual

Use when the user provides an earlier Collaboration Manual and wants it refreshed.

Read update-existing.md completely and follow it.

Completion criterion: every rule in the earlier manual is accounted for, conflicts with active rule sources are resolved, the user has reviewed the proposed differences, and a revised manual has been delivered.

Shared interview contract

  • Make every question skippable.
  • Infer operating traits from answers and real examples, not personality labels.
  • Distinguish can do from energized by; flag capable-but-draining work for support, process, delegation, or automation.
  • Distinguish stable preference, situational preference, task-scoped variable, and unknown.
  • Reflect observable patterns; ask the user to confirm interpretations.
  • Preserve the user's words when they are clearer than a framework label.
  • A deeper answer is an invitation signal, not consent to private exploration.
  • Never turn an inferred desire for depth into an automatic branch change.

Privacy gate

The AI-facing manual contains working instructions only. Apply the conversion rule:

Keep the how; remove the private why.

Before delivery, verify:

  • No family, relationship, medical, psychological, or other private history appears in the AI-facing manual.
  • No temporary emotion or current project fact is presented as a stable trait.
  • No MBTI or Holland label substitutes for observed behavior.
  • Uncertain judgments are marked as uncertain or omitted.

After delivery

If the user has not already made a persistence choice, read persistence.md and follow it. Do not ask the same persistence question twice.

If the completed Collaboration Manual leads the user to explicitly request private reflection, read self-manual.md. The private Self Manual is a separate-consent branch, not a default follow-up.

関連スキル