인터페이스는 현지화되어 있으며, Skill 소개와 설명 문서는 선택한 언어로 AI가 자동 번역합니다. 코드·명령어·경로·전문 용어는 원문을 유지하며 원작자 문서가 우선합니다.
Skills 둘러보기
생산성 및 협업 · SKILL.md

attune

attune

AI relational calibration — reading and adapting to the specific person you are working with. Goes beyond user-intent alignment (solving the right problem) to genuine attunement (meeting the person where they are). Maps communication style, expertise depth, emotional register, and implicit preferences from conversational evidence. Use at the start of a new …

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill 문서

GitHub

name: attune locale: caveman-ultra source_locale: en source_commit: 82c77053 fence_basis_commit: 82c77053 translator: "Julius Brussee homage — caveman" translation_date: "2026-04-19" description: > AI relational calibration — reading and adapting to the specific person you are working with. Goes beyond user-intent alignment (solving the right problem) to genuine attunement (meeting the person where they are). Maps communication style, expertise depth, emotional register, and implicit preferences from conversational evidence. Use at the start of a new session, when communication feels mismatched, after receiving unexpected feedback, or when transitioning between very different users or contexts. license: MIT allowed-tools: Read metadata: author: Philipp Thoss version: "1.0" domain: esoteric complexity: intermediate language: natural tags: esoteric, attunement, empathy, communication, calibration, meta-cognition, ai-self-application

Attune

Calibrate to person → read style, expertise, register, implicit prefs from conv evidence. Attunement > alignment: alignment = "right problem?"; attunement = "meet them where they are?"

Use When

  • New session start → calibrate before first substantive res
  • Comm mismatched → too formal/casual/long/sparse
  • Unexpected feedback → attunement gap
  • Big ctx shift (debug → brainstorm)
  • MEMORY.md holds user prefs
  • heal User-Intent check → surface OK but deeper off

In

  • Required: Current conv ctx (implicit)
  • Optional: MEMORY.md + CLAUDE.md via Read
  • Optional: Mismatch symptom ("explanations too long")

Do

Step 1: Receive — Gather Signals

Observe before adapt. Reception first, analysis second.

  1. Read msgs for how they communicate:
    • Length: short/direct or long/detailed?
    • Vocab: jargon, plain, mixed?
    • Tone: formal, casual, warm, efficient, playful?
    • Structure: lists, prose, bullets, stream?
    • Punctuation: precise, emoji, ellipses, exclam?
  2. Note what they skip → implicit assumptions
  3. MEMORY.md/CLAUDE.md → stored prefs = stable patterns

→ Comm fingerprint. Enough to match register.

If err: Signals ambiguous → match most recent msg tone. Attunement refines → not perfect immediately.

Step 2: Read — Assess Expertise and Context

Meet them at their level.

  1. Domain expertise:
    • Expert: precise terms, skips basics, nuanced Qs
    • Intermediate: knows concepts, asks specifics/edges
    • Beginner: foundational Qs, general language
  2. Tool familiarity:
    • High: refs tools/cmds/configs by name
    • Medium: knows want, not incantation
    • Low: outcome only, no tool refs
  3. Ctx depth:
    • Deep: carries implicit ctx
    • Moderate: knows project, not issue
    • Fresh: no prior ctx
Attunement Matrix:
┌──────────────┬──────────────────────────────────────────────────┐
│ Signal       │ Adaptation                                       │
├──────────────┼──────────────────────────────────────────────────┤
│ Expert       │ Skip explanations, use precise terms, focus on   │
│              │ the novel or non-obvious. They know the basics.  │
├──────────────┼──────────────────────────────────────────────────┤
│ Intermediate │ Brief context, then specifics. Confirm shared    │
│              │ understanding before going deep.                 │
├──────────────┼──────────────────────────────────────────────────┤
│ Beginner     │ Orient first, explain terms, provide context.    │
│              │ Don't assume; don't condescend.                  │
├──────────────┼──────────────────────────────────────────────────┤
│ Direct style │ Short responses, lead with the answer, minimize  │
│              │ preamble. Respect their time.                    │
├──────────────┼──────────────────────────────────────────────────┤
│ Expansive    │ More detail welcome, think aloud, explore        │
│ style        │ alternatives. They enjoy the journey.            │
├──────────────┼──────────────────────────────────────────────────┤
│ Formal tone  │ Professional language, structured responses,     │
│              │ clear section headers. Match their register.     │
├──────────────┼──────────────────────────────────────────────────┤
│ Casual tone  │ Conversational, contractions allowed, lighter    │
│              │ touch. Don't be stiff.                           │
└──────────────┴──────────────────────────────────────────────────┘

→ Clear sense of expertise + style from evidence, not stereotype.

If err: Hard to gauge → err slightly more ctx. Over-explain correctable; under-explain = user lost.

Step 3: Resonate — Match the Frequency

Resonance, not mimicry. Meet, not become.

  1. Match length: 2 sentences in → not 2 paragraphs out
  2. Match vocab: their terms. "function" → don't say "method" unless distinction matters
  3. Match structure: bullets → bullets, prose → prose
  4. Match energy: excited → engaged; frustrated → calm competence; exploratory → explore with
  5. Don't over-match: Matching ≠ flattening. Wrong → correct in their register

→ Noticeable quality shift. User feels heard. Response written for them.

If err: Matching feels forced → over-calibrating. Goal = natural resonance. Approximate fine.

Step 4: Sustain — Carry Attunement Forward

Attunement = ongoing practice, not one-time.

  1. After each msg → register shifted?
  2. Note when working (smooth) vs drifting (repeats, corrections)
  3. Explicit pref ("be concise") → overrides inference
  4. Stable pref worth preserving → MEMORY.md

→ Sustained quality + micro-adjustments as conv evolves.

If err: Attunement degrades long session → breathe → re-read user's most recent msg before res. Mid-session re-attune lighter than full cycle.

Check

  • Signals from actual evidence, not assumed
  • Expertise assessed w/ specific evidence
  • Res style matched register
  • Adaptation natural, not forced
  • Explicit prefs respected
  • Comm quality improved

Traps

  • Attunement as flattery: Match ≠ agree. Includes difficult truths in their register.
  • Over-calibrating: Too much effort on how → content suffers. Lightweight.
  • Expertise from identity: Don't infer from name/title/demographics. Read evidence.
  • Frozen calibration: Initial read = start. People shift. Keep reading.
  • Ignore explicit feedback: "too long" > any inference. Explicit beats implicit.

→

  • listen — deep attention for intent; attune = how, listen = what
  • heal — User-Intent Alignment check; attune goes deeper relational
  • observe — sustained neutral; attune applies it to person
  • shine — radiant authenticity; attune without it → mimicry
  • breathe — micro-reset → mid-session re-attune
저장한 Skills