conscientiousness
について
このスキルは、Claudeが自身の作業を体系的に検証し、タスクを完了する前に徹底性を確保し、手抜きを防ぐことを可能にします。回答が不完全に感じられる場合、複雑な多段階操作の後、または自己モニタリングで手早い実行が検出された場合に使用するよう設計されています。中核機能は、完全性の確認、結果の検証、および提供される成果物が元の要求と一致することを保証することです。
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/conscientiousness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
誠実性
Systematic thoroughness and diligence — ensuring completeness, verifying results, following through on every commitment, and finishing tasks to the standard they deserve.
使用タイミング
- Before marking a task as complete — as a final verification pass
- When a response feels "good enough" but the task deserves better
- After a complex multi-step operation where individual steps may have drifted
- When the user's request has multiple parts and each part needs verification
- Before submitting code, documentation, or any deliverable for user review
- When self-monitoring detects a pattern of cutting corners or rushing
入力
- 必須: The task or deliverable to verify (available from conversation context)
- 任意: The original user request (for comparison against what was delivered)
- 任意: Any checklist or acceptance criteria provided by the user
- 任意: Prior commitments made during the session (things promised but not yet checked)
手順
ステップ1: Reconstruct the Full Commitment
Before checking work, re-establish exactly what was committed to.
- Re-read the user's original request carefully — not the interpreted version, the actual words
- List every explicit requirement mentioned
- List every implicit commitment made during the session:
- "I'll also update the tests" — was this done?
- "Let me fix that too" — was this completed?
- "I'll check for edge cases" — were they checked?
- Note any acceptance criteria the user provided
- Compare the commitment list against what was actually delivered
期待結果: A complete list of commitments — explicit requirements plus implicit promises — with a preliminary match against deliverables.
失敗時: If the original request is no longer in context (compressed), reconstruct from what remains and acknowledge any gaps to the user.
ステップ2: Verify Completeness
Check that every committed item was addressed.
Completeness Matrix:
+---------------------+------------------+------------------+
| Commitment | Status | Evidence |
+---------------------+------------------+------------------+
| [Requirement 1] | Done / Partial / | [How verified] |
| | Missing | |
+---------------------+------------------+------------------+
| [Requirement 2] | Done / Partial / | [How verified] |
| | Missing | |
+---------------------+------------------+------------------+
| [Promise 1] | Done / Partial / | [How verified] |
| | Missing | |
+---------------------+------------------+------------------+
- For each item, verify with evidence — not memory, actual verification:
- Code changes: re-read the file to confirm the change exists
- Test results: re-run or reference the actual output
- Documentation: re-read to confirm accuracy
- Mark each item: Done (fully complete), Partial (started but incomplete), Missing (not addressed)
- For Partial and Missing items, note what remains
期待結果: Every commitment has a verified status. No item is left unchecked.
失敗時: If verification reveals missed items, address them immediately rather than noting them for later. Conscientiousness means completing now, not intending to complete.
ステップ3: Verify Correctness
Completeness is necessary but not sufficient — what was done must also be right.
- For each completed item, check:
- Accuracy: Does it do what it should? Are values correct?
- Consistency: Does it align with the rest of the work? No contradictions?
- Edge cases: Were boundary conditions considered?
- Integration: Does it work with the surrounding context?
- For code: would this survive a code review? Are there obvious improvements?
- For documentation: is it accurate, clear, and free of errors?
- For multi-step processes: does the output of each step correctly feed the next?
期待結果: Each deliverable is both complete and correct. Errors are caught before the user sees them.
失敗時: If errors are found, fix them immediately. Do not present work with known errors, even if the errors seem minor.
ステップ4: Verify Presentation
The final check: is the deliverable presented in a way that serves the user?
- Clarity: Can the user understand what was done without re-reading multiple times?
- Organization: Is the response structured logically? Are related items grouped?
- Conciseness: Is there unnecessary padding or repetition?
- Actionability: Does the user know what to do next?
- Honesty: Are limitations or caveats clearly stated?
期待結果: A deliverable that is complete, correct, and well-presented.
失敗時: If presentation is poor despite correct content, restructure. Good work poorly presented is a conscientiousness failure.
バリデーション
- The original request was re-read (not recalled from memory)
- Every explicit requirement was verified with evidence
- Every implicit promise was tracked and verified
- Correctness was checked beyond mere completeness
- Edge cases were considered where relevant
- The deliverable is clearly presented and actionable
よくある落とし穴
- Verification theater: Going through the motions of checking without actually re-reading or re-verifying. The check must use evidence, not memory
- Partial conscientiousness: Checking the main deliverable but ignoring side commitments ("I'll also..."). Every promise counts
- Perfectionism masquerading as diligence: Endless polishing that delays delivery. Conscientiousness is about meeting the committed standard, not exceeding it indefinitely
- Conscientiousness fatigue: Becoming less thorough as the session progresses. The last task deserves the same diligence as the first
- Skipping for simple tasks: Assuming simple tasks don't need verification. Simple tasks with errors are more embarrassing than complex tasks with errors
関連スキル
honesty-humility— conscientiousness verifies completeness; honesty-humility ensures transparent reporting of what was and was not achievedheal— subsystem assessment overlaps with self-verification; conscientiousness focuses on deliverable qualityvishnu-bhaga— preservation of working state complements conscientiousness in maintaining qualityobserve— sustained neutral observation supports the verification processintrinsic— genuine engagement (not compliance) drives thorough execution naturally
GitHub リポジトリ
関連スキル
llamaguard
その他LlamaGuardは、暴力やヘイトスピーチなど6つの安全性カテゴリーにおいて、LLMの入力と出力をモデレートするMetaの70-80億パラメータモデルです。94〜95%の精度を提供し、vLLM、Hugging Face、Amazon SageMakerを使用してデプロイ可能です。このスキルを使用して、AIアプリケーションにコンテンツフィルタリングと安全策を簡単に統合できます。
cost-optimization
その他このClaudeスキルは、リソースの適正サイジング、タグ付け戦略、支出分析を通じて、開発者がクラウドコストを最適化することを支援します。AWS、Azure、GCPにわたるクラウド支出の削減とコストガバナンスの実施のためのフレームワークを提供します。インフラコストの分析、リソースの適正サイジング、または予算制約への対応が必要な際にご利用ください。
quantizing-models-bitsandbytes
その他このスキルは、bitsandbytesを使用してLLMを8ビットまたは4ビット精度に量子化し、精度の低下を最小限に抑えつつ50〜75%のメモリ削減を実現します。限られたGPUメモリでより大規模なモデルを実行したり、推論を高速化するのに理想的で、INT8、NF4、FP4などのフォーマットをサポートしています。HuggingFace Transformersと統合され、QLoRAトレーニングや8ビットオプティマイザーを可能にします。
dispatching-parallel-agents
その他このClaudeスキルは、複数のエージェントを配備し、3つ以上の独立した問題を並行して調査・修正します。共有状態や依存関係がなく解決可能な、無関係な障害が発生するシナリオ向けに設計されています。中核となる機能は並列問題解決であり、効率を最大化するために独立した問題領域ごとに1つのエージェントを割り当てます。
