observe-guidance
について
このスキルは、システム、パターン、または現象を分析するための体系的な観察手法をユーザーに指導します。介入前に理解を深めるため、中立的な注意、フィールドノート、パターン認識、構造化された報告を指導します。結論に飛びつくのではなく、証拠に基づいた分析が必要な場合の、デバッグ、研究、またはダイナミクスの調査にご活用ください。
クイックインストール
Claude Code
推奨npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/observe-guidanceこのコマンドをClaude Codeにコピー&ペーストしてスキルをインストールします
ドキュメント
Observe (Guidance)
Guide person in systematic observation of system, phenomenon, or pattern. AI acts as field study coach — frame target, prepare protocol, sustain neutral attention, record findings with field notes, analyze patterns, report observations with clear separation of data and interpretation.
When Use
- Person wants to understand system behavior before intervening (debug by observation rather than trial and error)
- Someone conducting research or gathering evidence, needs structured observation methodology
- Person keeps jumping to conclusions, needs discipline of observation before interpretation
- Someone preparing report requiring evidence-based findings, not opinions
- Person wants to understand team dynamics, user behavior, or process effectiveness through direct observation
- After
meditate-guidancecultivated sustained attention, person wants to direct it toward specific system
Inputs
- Required: What person wants to observe (system, process, behavior, codebase, team dynamic, natural phenomenon)
- Required: Why observing (debugging, research, audit, curiosity, improvement)
- Optional: Time available (single session vs. multi-day study)
- Optional: Prior attempts to understand system (what already tried)
- Optional: Specific questions or hypotheses to test
- Optional: Tools for recording (notebook, screen capture, logging, metrics)
Steps
Step 1: Frame — Define Observation Target
Help person set up clear, bounded frame.
- Ask what they want to observe: "What system or behavior trying to understand?"
- Help narrow scope: "What specific aspect of that system interests you most?"
- Identify purpose: understanding, debugging, improvement, evidence-gathering, pure curiosity
- Set boundaries: what in scope and what not (prevents endless expansion)
- They have hypothesis? State explicitly, then set aside — "We look for evidence both for and against"
- Choose stance:
- Naturalist: observe without interfering (best for understanding behavior)
- Controlled: change one variable, observe effect (best for debugging)
- Longitudinal: observe over time (best for detecting trends)
Got: Clear frame with defined target, scope, purpose, stance. Person knows what they look at and what not.
If fail: Person can't narrow focus ("I want to understand everything")? Help pick one entry point: "What is one behavior you find most confusing?" Already committed to conclusion ("just need to prove X")? Gently challenge: "What would we need to see to disprove that? Let's look for both."
Step 2: Prepare — Set Up Protocol
Help person establish systematic recording approach.
- Choose method based on observation type:
- Codebase/system: file paths, line numbers, timestamps, log entries
- Behavior/process: time-stamped notes with actor, action, context
- Team/communication: quotes, speaker IDs, non-verbal cues
- Natural/physical: sketches, measurements, environmental conditions
- Create simple template:
Field Notes Template:
┌─────────────┬────────────────────────────────────────────────────────┐
│ Timestamp │ When the observation occurred │
├─────────────┼────────────────────────────────────────────────────────┤
│ Observation │ What was seen/heard/measured (fact only) │
├─────────────┼────────────────────────────────────────────────────────┤
│ Context │ What was happening around the observation │
├─────────────┼────────────────────────────────────────────────────────┤
│ Reaction │ Observer's response (thoughts, emotions, surprises) │
├─────────────┼────────────────────────────────────────────────────────┤
│ Hypothesis │ Tentative interpretation (kept separate from fact) │
└─────────────┴────────────────────────────────────────────────────────┘
- Emphasize separation: "Observation row is fact. Hypothesis row is interpretation. Never mix."
- Set minimum count: "Aim for at least 10 observations before drawing conclusions"
- If applicable, set up monitoring tools: logging, metrics, screen recording
Got: Person has recording method ready, understands critical distinction between observation and interpretation. Feels prepared to begin.
If fail: Template feels too formal? Simplify: "Just write what you see, separately write what you think it means." Resist recording ("I'll remember")? Explain unrecorded observations subject to memory bias — writing makes observation more accurate.
Step 3: Observe — Practice Sustained Neutral Attention
Guide person through actual session.
- Remind them of stance: "You are naturalist studying new species. Do not interfere — just watch"
- First 5 minutes: encourage pure observation without recording — just attend
- After initial immersion: begin recording using template
- Coach neutral language: "Instead of 'system crashed,' try 'system stopped responding at 14:32 after processing 47th request'"
- Watch for interpretation creeping into observation: "That is interpretation — record in hypothesis row"
- Encourage noting surprises: "What surprised you? Surprises often contain most valuable data"
- Periodically check frame: "Still observing what you set out to, or has attention drifted?"
- They want to intervene? "Note what you want to change and why, but don't change yet — keep observing"
Got: Person generates at least 5-10 concrete observations with specific evidence. Experiences difference between observing and interpreting, finds it harder than expected to maintain neutral attention.
If fail: Keeps interpreting instead of observing? Try this exercise: "Describe what you see as if explaining to someone who has never seen this system. Only verifiable facts." Run out of things quickly? Looking too high level — guide to zoom in: timing, ordering, edge cases, exceptions.
Step 4: Record — Capture Findings with Field Notes
Help person organize raw observations into structured notes.
- Review recorded observations together
- Check completeness: each observation has enough context to be understood later?
- Check factual accuracy: statements verifiable, or contain hidden assumptions?
- Group similar observations: "Any patterns forming?"
- Note frequencies: how often did each pattern appear?
- Note absences: "What did you expect to see that wasn't there?"
- Help separate strong observations (clear evidence) from weak (ambiguous data)
Got: Set of organized field notes cleanly separating observation from interpretation. Detailed enough someone else could verify observations independently.
If fail: Notes too vague ("things seemed slow")? Help add specifics: "How slow? Compared to what? In which conditions?" Too detailed (recording everything)? Help identify which observations relate to original frame, which are noise.
Step 5: Analyze — Identify Patterns and Generate Hypotheses
Guide person from observations to structured analysis.
- Lay out observations, look for patterns:
- Repetition: "This happened multiple times — systematic?"
- Correlation: "X always happens with Y — related?"
- Sequence: "A always precedes B — could A cause B?"
- Absence: "X never happens in condition Z — why?"
- Anomaly: "Everything follows pattern P except this one case — what's different?"
- For each pattern, ask: "Alternative explanation?"
- Generate 2-3 hypotheses explaining major patterns
- Distinguish correlation and causation: "Observing A and B co-occur doesn't prove A causes B"
- Identify testable hypotheses, what test confirms/refutes them
- Note confidence levels: which well-supported, which speculative?
Got: Person moves from raw observations to structured hypotheses while maintaining discipline of separating data from theory. Has at least one testable hypothesis for original question.
If fail: Jumps to single explanation immediately? Challenge: "That's one possibility. What's another?" Sees no patterns? Observations may be too few — continue observation before analysis. Every observation points to same conclusion? May be filtering — ask: "What evidence would contradict your current theory?"
Step 6: Report — Share Findings with Clear Structure
Help person communicate observations effectively.
- Structure report:
- Context: What observed, when, why, under what conditions
- Method: How observation conducted (protocol, tools, duration)
- Findings: Key observations with evidence (data, not interpretation)
- Analysis: Patterns identified, hypotheses generated, confidence levels
- Recommendations: Next steps (further observation, testing, intervention)
- Limitations: What observation did not cover, potential biases
- Help write findings in neutral language separating fact from interpretation
- Review for hidden assumptions or unsupported claims
- Observations for debugging? Translate hypotheses into concrete tests
- Observations for report? Ensure evidence cited specifically
- Observations for personal understanding? Summarize key insights and remaining questions
Got: Clear report communicating observations, patterns, hypotheses while maintaining distinction between what observed and what inferred. Reader can evaluate evidence independently.
If fail: Report buries observations in interpretation? Restructure: "Put all facts in one section, all theories in another." Lacks confidence levels ("this is definitely because...")? Help calibrate: "How sure? What would change your mind?"
Checks
- Frame set before observation began (not free-form wandering)
- Recording protocol established and used consistently
- Observations recorded as facts, separate from interpretations
- At least 5 concrete, evidence-backed observations captured
- Patterns identified through analysis, not assumed from start
- Hypotheses testable, with stated confidence levels
- Person experienced discipline of observing before interpreting
Pitfalls
- Observation as confirmation bias: Observing only things supporting pre-existing belief. Frame should include "look for evidence against your hypothesis" as explicit instruction
- Intervention urge: Seeing problem and wanting to fix immediately. Premature intervention masks root cause — observe first, then intervene with full understanding
- Recording fatigue: Detailed observation mentally taxing. Suggest breaks and realistic session lengths (30-60 min focused observation is substantial)
- Overcomplicating protocol: For simple observations, notebook and timestamps enough. Protocol should serve observation, not replace it
- Confusing observation with surveillance: In interpersonal observation, ethical boundaries matter. Observe visible behavior, don't spy. If observing people, transparency usually better than secrecy
- Skipping frame: Without clear target, attention scatters, findings unfocused. Even rough frame better than none
See Also
observe— AI self-directed variant for sustained neutral pattern recognition across systemslearn-guidance— observation feeds learning by providing raw data for understandinglisten-guidance— listening is focused observation of speaker; observation broader-scope attention to any systemremote-viewing-guidance— shares structured observation methodology adapted for non-local perceptionread-garden— garden observation skill using similar CRV-adapted sensory protocols
GitHub リポジトリ
関連スキル
evaluating-llms-harness
テストこのClaudeスキルは、lm-evaluation-harnessを実行し、MMLUやGSM8Kなど60以上の標準化学術タスクでLLMをベンチマークします。開発者がモデルの品質を比較し、トレーニングの進捗を追跡し、学術的な結果を報告するために設計されています。このツールはHuggingFaceやvLLMモデルを含む様々なバックエンドをサポートしています。
cloudflare-cron-triggers
テストこのスキルは、cron式を使用してWorkersをスケジュールするためのCloudflare Cron Triggersの実装に関する包括的な知識を提供します。定期的なタスクの設定、メンテナンスジョブ、自動化されたワークフローの構築を網羅し、無効なcron式やタイムゾーン問題といった一般的な課題への対処法も含みます。開発者はこれを使用して、スケジュールされたハンドラーの設定、cronトリガーのテスト、WorkflowsやGreen Computeとの連携を構成できます。
webapp-testing
テストこのClaude Skillは、Playwrightベースのツールキットを提供し、Pythonスクリプトを通じてローカルWebアプリケーションのテストを可能にします。フロントエンドの検証、UIデバッグ、スクリーンショット撮影、ログ表示を実現し、サーバーライフサイクルを管理します。ブラウザ自動化タスクにご利用いただけますが、コンテキストの汚染を避けるため、スクリプトのソースコードを読むのではなく直接実行してください。
finishing-a-development-branch
テストこのスキルは、開発者がテストの合格を確認し、構造化された統合オプションを提示することで、完成した作業を仕上げることを支援します。実装が完了した後のマージ、PR作成、ブランチの整理といったワークフローを案内します。コードが準備できてテスト済みの際に使用し、開発プロセスを体系的に完了させましょう。
