MCP HubMCP Hub
스킬 목록으로 돌아가기

expand-awareness

pjt222
업데이트됨 2 days ago
7 조회
17
2
17
GitHub에서 보기
기타ai

정보

이 스킬은 집중된 단일 영역 주의에서 광범위한 다중 영역 인식으로 전환하여 영역 간 연결을 파악합니다. 여러 영역에 걸친 문제를 다루거나 집중적 분석이 더 넓은 패턴을 놓칠 때 유용합니다. 이 접근 방식은 좁은 스포트라이트보다는 영역 전체에 걸친 인지적 '방송'과 유사합니다.

빠른 설치

Claude Code

추천
기본
npx skills add pjt222/agent-almanac -a claude-code
플러그인 명령대체
/plugin add https://github.com/pjt222/agent-almanac
Git 클론대체
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/expand-awareness

Claude Code에서 이 명령을 복사하여 붙여넣어 스킬을 설치하세요

문서

Expand Awareness

Open attention from one-domain focus to wide-view many-domain awareness. Where meditate clears noise to sharpen focus, and observe watches one thing close, expand-awareness deliberate widens aperture to hold all relevant domains in view at once — not analyzing any one, but seeing all at once.

Three traditions feed this practice. Baars' Global Workspace Theory models mind as broadcast — once info enters global workspace, it opens to all cognitive processes at once rather than run in one specialist module. The dzogchen idea of rigpa shows open, unfocused awareness that sees whole field with no fix on any part of it. And jazz ensemble listening needs each player to hear all players at once — not tracking one instrument then another, but holding full sound field so their own part fits the whole.

Common thread: wide-view perception is not sum of many focused perceptions. It is a different kind of attention that shows patterns invisible to any single focused view. The links between domains — tensions, echoes, gaps — only become visible when all domains held in view at once.

When Use

  • After meditate has cleared context noise — cleared space is canvas for open
  • Problem spans many domains and single-domain analysis keeps missing cross-domain links
  • Before integrate-gestalt — open produces raw wide-view perception that gestalt then folds
  • When narrow to one domain keeps failing to fix problem that lives between domains
  • When you notice yourself cycling through domains in sequence and want to hold them at once instead
  • At start of complex project to see its full scope before planning starts
  • When solution that works perfect in one domain makes problems in another — cross-domain tension needs to be seen whole, not debugged domain by domain

Inputs

  • Required: Problem or situation that touches many domains
  • Optional: First domain list (if relevant domains already known; Step 1 will refine)
  • Optional: Specific cross-domain tension or question to hold during open (e.g., "Why does testing strategy clash with deployment arch?")
  • Optional: Old open notes from prior session on same problem (enables deepen over time)

Steps

Five steps trace natural arc: know what to see (Inventory), drop narrow default (Soften), open to whole field (Expand), learn to stay open (Sustain), catch what was seen (Note). Steps 3 and 4 are core practice; steps 1, 2, 5 are prep and keeping.

Full sequence can take as little as few minutes for known problem or much longer for novel multi-domain challenge.

When used as part of synoptic cycle (meditate -> expand-awareness -> observe -> integrate-gestalt -> express-insight), this skill holds second spot: cleared space from meditation becomes canvas for open.

Skill can also be used alone — any time multi-domain seeing would help, no matter if full synoptic cycle is called.

Step 1: Inventory — Enumerate the Domains

Map all domains tied to current problem. This looks like what polymath does when break down problem — but the key difference is polymath hands each domain to specialist. Here, no handoff. Point is not divide and conquer but know what wide-view should hold, so open has content to hold.

  1. Name each domain that touches current problem. Be specific: not "code" but "R package structure," not "infrastructure" but "GitHub Actions CI/CD"
  2. For each domain, state in one sentence what view it brings to problem — what does this domain see that others do not?
  3. Note which domains are comfortable (well-known) and which are edge (relevant but less known). Edge domains often hold key insight just because under-attended. Mark comfort level for each: strong, moderate, edge
  4. Look for domains not obvious: human factors side? upkeep side? time side (how this shifts over time)? political or org side?
  5. If can, use Read, Grep, or Glob to scan real workspace — file structure, config files, and docs often show domains pure thinking misses
  6. Aim for 3-7 domains. Fewer than 3 hints problem may not need wide-view awareness. More than 7 risks spread rather than open

Got: List of 3-7 domains, each with its view on problem. List feels full — not exhaustive (every possible domain) but enough (every domain that materially hits problem). Each domain named specific enough you could hand to specialist and they would know their scope.

If fail: Only one domain shows? Problem may truly be single-domain — use observe instead. Only two domains show? Think whether there is bridging domain between them (there usually is — it's space where two interact). Dozens of domains show? Group related into clusters and treat each cluster as one domain for open. Goal is at-once perception, not full taxonomy.

Step 2: Soften Focus — Release Single-Domain Concentration

Move from focused attention on one domain to diffuse ready to see all domains. This is different from meditate — meditation clears noise (distract thoughts, left-over context, feel residue), while this step clears narrowness (habit of focus on single domain to block of others). Goal is not empty but open.

  1. Notice where attention is anchored now. Which domain is it drawn to? Name it clear: "My attention is anchored on testing domain"
  2. Gently drop that anchor. Not by forcing attention away, but by loosening grip — let domain move from front to middle ground. It stays but no longer rules
  3. Do not swap it with another domain. Let attention float with no target. This is uncomfortable moment — gap between drop one focus and not yet having another
  4. If attention snaps back to anchor domain, notice pull without fight. Each notice loosens grip more. Pull itself holds info about what feels most urgent
  5. Notice quality of attention itself:
    • Focused attention feels sharp and narrow, like spotlight beam
    • Softened attention feels wide and ambient, like daylight filling room
    • Aim for daylight — open, even, not aimed at any single point
  6. Move feels like relaxing clenched fist — hand does not vanish, it just opens

Got: Mind state where no single domain rules attention. Mind is open and receptive rather than focused and aimed. This state is unfamiliar and slightly uncomfortable — that discomfort is signal narrowness was released. There is urge to fill openness fast; resist it.

If fail: Focus won't soften — one domain keeps demanding attention? It may hold unresolved urgency. Deal with it brief (note urgent item, promise to return) and then try soften again. Analytic mind protests this "unproductive"? Note protest as itself form of narrowness: habit of always needing single target is just the habit this step drops.

Step 3: Expand — Hold All Domains Simultaneously

Deliberate bring all listed domains into awareness at once. Not by thinking about each in sequence, but by seeing them as single field — way you see whole landscape rather than scan tree by tree.

This is Global Workspace moment: info that was siloed in specialist modules is now broadcast to all at once. In jazz terms, this is moment when musician stops tracking single instruments and starts hearing whole ensemble as one sound.

  1. Start with domain that feels most natural or central. Bring into awareness — not as problem to solve, but as view to hold. Let it be present without analysis
  2. Without drop first, bring second domain next to it. Hold both. This is first open — from one to two. Notice if two domains touch right away or sit apart
  3. Keep adding domains until all present at once. This is core act of skill — it needs holding without grasping, presence without analysis
  4. Notice what shows up between domains:
    • Connections: where do two domains share idea, limit, or pattern?
    • Tensions: where do two domains pull opposite?
    • Resonances: where does pattern in one domain echo unexpected in another?
    • Gaps: where is empty space between domains that nothing covers?
    • Surprises: what is visible from wide-view that was invisible from any single domain?
  5. Do not chase any single link. Let them log without analysis. Wide-view is the product, not any single insight inside
  6. If using tools, this is moment to read files from many domains quick after each other — not to analyze one by one, but to let them sit in context

Got: Felt sense of holding many views at once. Links, tensions, echoes show with no force. Experience more like seeing pattern in mosaic than reading list of items. Between-domain space — where no single domain rules — becomes visible. This is where new insights live: not inside any domain but in ties between them.

If fail: Domains keep falling into sequence attention (think A, then B, then C)? Try spatial metaphor: place each domain at different spot in imagined space and "look" at whole space rather than at any single spot. Number of domains swamps? Drop to 3 most central and open from there. Feel purely mental and detached? Ground it: for each domain, touch one concrete artifact (file, config, test) and then re-open.

Step 4: Sustain — Maintain the Panoramic View

Wide-view is unstable. Attention naturally narrows back to single domain — this not fail but nature of focused thinking reasserting itself. This step teaches holding, not acting. Goal is to sustain wide view long enough for cross-domain patterns to become visible.

  1. Hold opened awareness from Step 3. Notice when it starts to shrink
  2. When attention narrows to one domain — and it will — notice narrowing without judge. Name which domain pulled: "Attention narrowed to security" or "Attention narrowed to testing." Naming matters; it makes unconscious narrowing conscious
  3. Re-open gentle. Do not force wide view; invite it back by recalling full inventory from Step 1. Touch each domain light, like re-tuning to each instrument in ensemble
  4. Each cycle of narrow-notice-re-open strengthens capacity to sustain. Cycles are the practice, not break of practice. Three cycles is minimum; five is better
  5. Pay special heed to which domain keeps pulling focus. This pull is info — it may show where problem gravity center lives, or where unresolved anxiety lumps
  6. Notice whether wide-view shifts between cycles:
    • New patterns often show on second or third open that were not visible in first
    • Ties between domains may shift as you hold them longer
    • What looked like tension may show itself as creative constraint
  7. Sustain at least 2-3 cycles of narrow and re-open before go on

Got: Growing power to hold wide-view for longer times. Narrow-notice-re-open cycle becomes smoother with each repeat. Domains that keep pulling focus get marked as potential gravity centers. By second or third cycle, wide-view may start feeling natural rather than effortful — this is signal capacity is growing.

If fail: Wide-view won't sustain at all — attention falls fast every time? Drop count of domains. Start with two, sustain that, then add third. Capacity grows step by step, not all at once. Sustained open gives anxiety or sense of losing grip? Ground by briefly touch one concrete detail (file path, function name, specific fact) and then re-open from that grounded spot. One domain keeps hijacking attention? It may need focused work before open can hold it — handle urgency first, then return to wide-view.

Step 5: Note — Record Before Narrowing

Wide-view is temporary by nature. Before let attention narrow back to focus mode, catch what was seen from wide view. These notes are raw stuff for integrate-gestalt and they fade fast — what is sharp during open becomes vague once focus returns. Record now.

  1. Write down every link, tension, echo, and gap noticed during open — even faint ones. Faint perceptions often prove most key once integrate-gestalt grows them
  2. Note which domains kept pulling focus and what that pull might mean
  3. Note any surprises — perceptions that would not come from examining any single domain alone. These cross-domain insights are unique product of wide awareness
  4. Note overall shape of problem as seen from wide view. Does it look different from how it looked in single-domain focus? Has problem gravity center shifted?
  5. Note quality of open itself: was it easy to sustain? which domains resisted inclusion? how many cycles of narrow-and-re-open? This meta-watch helps future open sessions
  6. If this open was part of synoptic cycle, tag notes clear for integrate-gestalt to consume
  7. Do not analyze or rank notes yet. Raw perception first; synthesis comes later (in integrate-gestalt)
  8. Let attention narrow natural. Open is done. Do not force more open — wide view has served its purpose

Got: Record of cross-domain perceptions: links, tensions, echoes, gaps, surprises. Record catches what was visible only from wide view — insights single-domain analysis would miss. Move back to focus attention feels natural, not forced. Notes are raw stuff, not polished conclusions.

If fail: No cross-domain perceptions came? Domains may be more alone than assumed — which is itself finding worth logging ("these domains do not interact" is valuable know). Too many perceptions came to catch? Record 3-5 strongest and note that more exist. Completeness is not goal; catching wide-view signal is. Notes feel too abstract? Anchor each in concrete artifact: "Tension between API design and security model is visible in auth middleware at X."

Validation

  • At least 3 domains listed with their specific views on problem
  • Focus was softened from single-domain tight before try open
  • Many domains held in awareness at once, not run in sequence
  • At least one cross-domain link, tension, or echo was seen during open
  • Wide-view sustained through at least one cycle of narrow and re-open
  • Perceptions logged before attention narrowed, keeping wide-view signal
  • Logged notes tell between what was seen (raw watch) and what it might mean (read)
  • Open gave at least one insight that would not have come from sequence single-domain analysis

Pitfalls

  • Sequence scan instead of at-once see: Fast cycle through domains one by one is analysis, not wide awareness. Split is between seeing landscape and reading list of its features. If you find yourself thinking "first A, then B, then C," you are scanning, not opening
  • Mix open with diffuse: Opened awareness is alert and receptive — you can feel each domain clear even while holding all. Diffuse attention is scattered and unfocused — everything blurs. If everything feels blurry, attention has diffused rather than opened. Re-anchor brief on one domain and open again from that grounded spot
  • Analyze during open: Urge to right away chase link collapses wide-view into focus attention on that one link. Note it and keep holding wide view. Analysis has its turn later, in integrate-gestalt
  • Skip inventory: Open without knowing what to open into gives vague space rather than wide perception. Inventory gives content awareness opens to hold. Without it, "open" has no direction
  • Rush through soften: Move from focus to open attention takes time. Skip Step 2 means try open from still-focused state, which gives sequence scan dressed as at-once see
  • Force links: Not all domains link. Make links between truly alone domains pollutes perception. Let links show up or not. Absence of link is data, not fail
  • Open without prior clear: Open from noisy base amplifies noise. Run meditate first when context noise present
  • Treat open as one-time event: Wide awareness is practice that deepens with repeat. First open session shows surface links; later sessions on same problem show structural patterns. Return to this skill as problem grows
  • Mix open with expert: Hold many domains in awareness does not make you expert in all. Open shows where to look, not what to conclude. Deep single-domain work still needed after open spots critical crossings

See Also

  • meditate — clear space that open fills; meditate before open for best
  • integrate-gestalt — takes raw perceptions from open and folds into coherent whole
  • observe — focused watch of single target; open is wide-field watch of many targets at once
  • dream — unconstrained link roam; open is disciplined wide perception, not free link
  • remote-viewing — enter problems with no mask; shares open openness but enters unknown ground rather than known multi-domain space
  • breathe — micro-pause that can help reset when open collapses; single conscious breath before re-open
  • heal — if open keeps showing same unresolved tension, deeper self-check of heal may be needed before more open
  • adaptic — meta-skill composing full synoptic cycle, where expand-awareness is second move

GitHub 저장소

pjt222/agent-almanac
경로: i18n/caveman/skills/expand-awareness
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

연관 스킬

llamaguard

기타

LlamaGuard는 폭력 및 혐오 발언 등 6가지 안전 범주에서 LLM 입력과 출력을 조정하기 위한 Meta의 70-80억 파라미터 모델입니다. 94-95% 정확도를 제공하며 vLLM, Hugging Face 또는 Amazon SageMaker를 사용해 배포할 수 있습니다. 이 기술을 사용하여 AI 애플리케이션에 콘텐츠 필터링 및 안전 가드레일을 손쉽게 통합하세요.

스킬 보기

cost-optimization

기타

이 Claude Skill은 리소스 적정화, 태깅 전략, 지출 분석을 통해 개발자들이 클라우드 비용을 최적화할 수 있도록 지원합니다. AWS, Azure, GCP에서 클라우드 비용을 절감하고 비용 거버넌스를 구현하기 위한 프레임워크를 제공합니다. 인프라 비용을 분석하거나, 리소스를 적정화하거나, 예산 제약을 충족해야 할 때 사용하세요.

스킬 보기

quantizing-models-bitsandbytes

기타

이 스킬은 bitsandbytes를 사용하여 LLM을 8비트 또는 4비트 정밀도로 양자화하며, 최소한의 정확도 손실로 50-75%의 메모리 감소를 달성합니다. 제한된 GPU 메모리에서 더 큰 모델을 실행하거나 추론을 가속화하는 데 이상적이며, INT8, NF4, FP4와 같은 형식을 지원합니다. 이 스킬은 HuggingFace Transformers와 통합되어 QLoRA 학습 및 8비트 옵티마이저를 가능하게 합니다.

스킬 보기

dispatching-parallel-agents

기타

이 Claude Skill은 3개 이상의 독립적인 문제를 동시에 조사하고 해결하기 위해 다중 에이전트를 배치합니다. 공유 상태나 의존성 없이 해결 가능한 무관련 장애 시나리오에 맞게 설계되었습니다. 핵심 기능은 병렬 문제 해결로, 각 독립 문제 영역마다 하나의 에이전트를 할당하여 효율성을 극대화합니다.

스킬 보기