remote-viewing
について
このスキルは、構造化された調査プロトコルを適応させることで、未知のコードベースや問題の体系的な探索を可能にします。段階的なデータ収集と仮説管理を通じて、早急な結論を避けるよう導きます。複雑なシステムのデバッグや、過去の仮定が誤りを招いた未知の領域に取り組む際に活用してください。
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/remote-viewing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
Remote View
Approach unknown codebase/problem/system using CRV protocol adapted for AI investigation — gather raw obs before conclusions, manage premature labeling (AOL), build understanding via staged data collection.
Use When
- Investigate unfamiliar codebase, arch unknown
- Debug problem, root cause not obvious + premature hypotheses could mislead
- Explore domain/tech w/ limited ctx
- Prev attempts led astray by assumptions
- Any problem where "beginner's mind" > pattern matching
In
- Required: Target (codebase path, problem desc, system to understand)
- Required: Commitment to blind approach — resist conclusions til collection complete
- Optional: Specific questions about target (save Stage V)
- Optional: Prior meditation for assumption-clearing (see
meditate)
Do
Step 1: Cooldown — Clear Assumptions
Transition assumption-heavy → receptive obs. Non-negotiable.
- ID all preconceptions about target:
- "This is probably a React app" — declare
- "The bug is likely in the database layer" — declare
- "This follows MVC architecture" — declare
- Write each preconception explicit (in reasoning or output)
- For each: "This may or may not be true. I will verify, not assume."
- Release need to ID target quickly — goal = accurate description, not fast labeling
- When notice analytical mind reaching for framework/label, pause + redirect to raw obs
→ List of declared preconceptions + conscious shift "I think I know" → "I will observe what this actually is." Alert + receptive, not jumping conclusions.
If err: assumptions keep reasserting ("but it really IS a React app...") → extend cooldown. Write to "parking lot" + continue. Don't begin gathering while attached to specific hypothesis — colors everything.
Step 2: Ideogram — First Contact (Stage I)
Initial contact w/ target via most minimal obs possible.
- Use
Glob→ top-level structure only (e.g.*orpath/*) — no read files yet - Note immediate unfiltered impressions: file count, naming patterns, presence/absence of obvious markers
- Record raw obs w/ simple descriptors:
- "many small files" not "microservice architecture"
- "deeply nested directories" not "enterprise Java"
- "single large file" not "monolith"
- Decode initial impression into 2 components:
- A (activity): Active or dormant? Growing or stable? Simple or complex?
- B (feeling): Organized or chaotic? Dense or sparse? Familiar or alien?
- Write A + B assessments — first data points
→ Handful of raw low-level obs about target's surface. No names, labels, architectural patterns — just shapes, sizes, textures.
If err: immediately categorize ("oh, Next.js app") → declare AOL (Step 6), extract raw descriptors underneath label ("JavaScript files, nested pages directory, package.json present"), continue w/ raw.
Step 3: Sensory — Raw Data (Stage II)
Systematically collect raw data w/o interpretation.
Stage II Data Channels for Codebase Investigation:
┌──────────────────┬────────────────────────────────────────────────────┐
│ Channel │ What to Observe │
├──────────────────┼────────────────────────────────────────────────────┤
│ File patterns │ Extensions, naming conventions, file sizes │
│ │ (NOT frameworks — just patterns) │
├──────────────────┼────────────────────────────────────────────────────┤
│ Directory shape │ Depth, breadth, nesting patterns, symmetry │
├──────────────────┼────────────────────────────────────────────────────┤
│ Configuration │ What config files exist? How many? What formats? │
├──────────────────┼────────────────────────────────────────────────────┤
│ Dependencies │ Lock files present? How large? How many entries? │
├──────────────────┼────────────────────────────────────────────────────┤
│ Documentation │ README present? How long? Other docs? Comments? │
├──────────────────┼────────────────────────────────────────────────────┤
│ Test presence │ Test directories? Test files? Ratio to source? │
├──────────────────┼────────────────────────────────────────────────────┤
│ History signals │ Presence of .git/, CHANGELOG/RELEASE_NOTES, │
│ │ lockfile timestamps (via Glob/Read if accessible) │
├──────────────────┼────────────────────────────────────────────────────┤
│ Energy/activity │ Which areas changed recently? Which are dormant? │
└──────────────────┴────────────────────────────────────────────────────┘
- Probe each channel via
Glob,Grep, lightRead - Record one obs per channel — first impression, no deep-dive
- Use descriptive terms not labels: "73 .ts files" not "TypeScript project"
- Circle (mark) any obs feeling significant
- Channel produces nothing → record "nothing observed" + move on
- Aim 10-20 data points across channels
→ List of raw obs feeling discovered not assumed. Some significant, some noise. Low-level descriptions, not high-level categorizations.
If err: every obs becomes categorization → slipped into analysis. Stop, return ideogram, re-contact w/ fresh eyes. One channel dominates (all file obs, no history) → deliberately shift to underused.
Step 4: Dimensional — Structure (Stage III)
Move raw obs → spatial + structural understanding.
- Begin mapping target arch w/o labeling:
- What connects to what? (imports, refs, config pointers)
- Major "areas" + how relate?
- Hierarchy — flat, nested, mixed?
- Read few key files lightly — entry points, configs, README
- Note relationships: "directory A imports from directory B," "config file references paths in C"
- Sketch spatial layout: how does info flow through system?
- Record Aesthetic Impact (AI) — how does codebase feel? Well-maintained? Rushed? Experimental?
→ Rough structural map w/ relationship annotations. General scope (large/small, simple/complex, monolithic/modular) clearer. "Feeling" of codebase captured.
If err: map feels pure guesswork → simplify: note only verifiable connections (actual imports, actual config refs). No structural patterns emerge → return Stage II + collect more raw data — dimensional needs foundation.
Step 5: Interrogation — Directed Q (Stage V)
Classic CRV Stage IV → deeper analytical structure; for codebase investigation that work intentionally merged into earlier dimensional/structural stages above, so this proceeds to Stage V for directed q.
Now, only now, bring specific q to investigation.
- State each q explicit: "What is entry point?" "Where does data come from?" "Test coverage?"
- For each q, search via
Grep+Read— targeted not exploratory - Record first finding for each q
- Note confidence: high (direct evidence), medium (inferred), low (uncertain)
- Mark all Stage V data clearly — higher AOL risk because q prime expectations
→ Specific answers to directed q, grounded in raw + structural data already collected. Confidence levels honest.
If err: directed q produce only AOL (answering from assumption not evidence) → return earlier stages. Protocol sequential for reason — skipping obs + jumping q → unreliable answers.
Step 6: Manage AOL
AOL = primary source of error. Occurs when analytical mind prematurely labels target. Manage entire session.
AOL Types in Codebase Investigation:
┌──────────────────┬─────────────────────────────────────────────────┐
│ Type │ Description and Response │
├──────────────────┼─────────────────────────────────────────────────┤
│ AOL (labeling) │ "This is a Django app" — Declare: "AOL: Django"│
│ │ Extract raw descriptors: "Python files, urls.py,│
│ │ migrations directory, settings module." │
├──────────────────┼─────────────────────────────────────────────────┤
│ AOL Drive │ The label becomes insistent: "This HAS to be │
│ │ Django." Declare "AOL Drive" and pause. What │
│ │ evidence contradicts the label? Look for it. │
├──────────────────┼─────────────────────────────────────────────────┤
│ AOL Signal │ The label may contain valid information. After │
│ │ declaring, extract: "Django" → "URL routing, │
│ │ ORM pattern, middleware chain." These raw │
│ │ descriptors are valid data even if "Django" is │
│ │ wrong. │
├──────────────────┼─────────────────────────────────────────────────┤
│ AOL Peacocking │ An elaborate narrative: "This was built by a │
│ │ team that was migrating from Java and..." This │
│ │ is imagination, not signal. Declare "AOL/P" and │
│ │ return to raw observation. │
└──────────────────┴─────────────────────────────────────────────────┘
Discipline ≠ avoiding AOL → recognizing + declaring so no contaminate. Every investigation produces AOL. Skill = how fast you catch.
→ AOL recognized in moments, declared explicit, investigation continues w/ raw descriptors not labels.
If err: AOL has taken over (reasoning from label for several steps) → call "AOL Break". Return Stage II + collect new raw obs testing label. Heavily contaminated investigation noted as such in review.
Step 7: Close + Review
End investigation formally + synthesize findings.
- Review all data in order: first impressions, raw obs, structural data, directed answers, AOL declarations
- ID 5-10 obs w/ highest confidence
- Now — only now — form synthesis: what is this system? how works? key characteristics?
- Note which parts well-supported by evidence vs inferred
- Compare synthesis vs preconceptions declared Step 1 — confirmed? wrong?
- Document findings for user or own future ref
→ Grounded understanding built up from raw obs not assumed from pattern matching. Synthesis more accurate than quick categorization, confidence levels honest.
If err: synthesis feels thin → earlier stages may not collected enough. Don't dismiss partial findings — description of "73 TypeScript files, deeply nested component structure, active git history, thin test coverage" > wrong label. Accurate description = goal not identification.
Check
- Preconceptions declared before collection
- Stage I obs = raw descriptors not labels
- Stage II data collected across multi channels not just one
- All AOL declared at moment of recognition
- Stages progressed sequential (I → II → III → V), no jumping
- Target approached blind — no files read on assumptions
- Synthesis distinguishes evidence-supported from inferences
- Investigation record preserved for future ref
Traps
- Jump to ID: Searching "what framework?" before raw obs guarantees AOL contamination
- Suppress labels: Trying not to form hypotheses creates tension → declare them + extract raw signal underneath
- Skip cooldown: Start investigation while attached to hypothesis biases all obs
- Confirmation-only search: Once hypothesis forms, search only confirming + ignore contradictions
- Confuse speed w/ skill: Fast ID feels productive but often wrong. Thorough staged obs longer but more accurate
- Insufficient channel diversity: Investigating only one lens (only code, only structure) misses signals via other channels
→
remote-viewing-guidance— human-guidance variant where AI = CRV monitor/taskermeditate— mental stillness + assumption-clearing in meditation directly improves investigation qualityheal— when investigation reveals AI's own reasoning biases, self-healing addresses root cause
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。
