スキル一覧に戻る

coordinate-reasoning

pjt222
更新日 Yesterday
3 閲覧
17
2
17
GitHubで表示
コミュニケーションai

について

このスキルは、複雑な多段階タスクにおいてClaude自身の内部推論を調整するために、コンテキストとメモリ内の情報の鮮度と減衰を管理します。副次タスクが相互に連携する必要がある場合、単純な局所プロトコルである「スティグマージック信号」を用いて、一貫した動作を維持します。長時間のタスク中、コンテキスト圧縮後、または情報の陳腐化が連携を損なう可能性がある場合にご利用ください。

クイックインストール

Claude Code

推奨
メイン
npx skills add pjt222/agent-almanac -a claude-code
プラグインコマンド代替
/plugin add https://github.com/pjt222/agent-almanac
Git クローン代替
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/coordinate-reasoning

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

Coordinate Reasoning

Manage internal coordination of reasoning processes using stigmergic principles — treat context as environment where information signals have freshness, decay rates, and interaction rules that produce coherent behavior from simple local protocols.

When Use

  • During complex tasks where multiple sub-tasks must coordinate (multi-file edits, multi-step refactoring)
  • Context grown long, information freshness uncertain
  • After context compression when some information may have been lost
  • Sub-task outputs need to feed into each other cleanly
  • Earlier reasoning results need to be carried forward without degradation
  • Complementing forage-solutions (exploration) and build-coherence (decision) with execution coordination

Inputs

  • Required: Current task decomposition (what sub-tasks exist, how do they relate?)
  • Optional: Known information freshness concerns (e.g., "I read that file 20 messages ago")
  • Optional: Sub-task dependency map (which sub-tasks feed into which?)
  • Optional: Available coordination tools (MEMORY.md, task list, inline notes)

Steps

Step 1: Classify Coordination Problem

Different coordination challenges require different signal designs.

AI Coordination Problem Types:
┌─────────────────────┬──────────────────────────────────────────────────┐
│ Type                │ Characteristics                                  │
├─────────────────────┼──────────────────────────────────────────────────┤
│ Foraging            │ Multiple independent searches running in         │
│ (scattered search)  │ parallel or sequence. Coordination need: share   │
│                     │ findings, avoid duplicate work, converge on      │
│                     │ best trail                                       │
├─────────────────────┼──────────────────────────────────────────────────┤
│ Consensus           │ Multiple approaches evaluated, one must be       │
│ (competing paths)   │ selected. Coordination need: independent         │
│                     │ evaluation, unbiased comparison, commitment      │
├─────────────────────┼──────────────────────────────────────────────────┤
│ Construction        │ Building a complex output incrementally (multi-  │
│ (incremental build) │ file edit, long document). Coordination need:    │
│                     │ consistency across parts, progress tracking,     │
│                     │ dependency ordering                              │
├─────────────────────┼──────────────────────────────────────────────────┤
│ Defense             │ Maintaining quality under pressure (tight time,  │
│ (quality under      │ complex requirements). Coordination need:        │
│ pressure)           │ monitoring for errors, rapid correction,         │
│                     │ awareness of degradation                         │
├─────────────────────┼──────────────────────────────────────────────────┤
│ Division of labor   │ Task decomposed into sub-tasks with              │
│ (sub-task mgmt)     │ dependencies. Coordination need: ordering,       │
│                     │ handoff, result integration                      │
└─────────────────────┴──────────────────────────────────────────────────┘

Classify current task. Most complex tasks are Construction or Division of Labor; most debugging tasks are Foraging; most design decisions are Consensus.

Got: Clear classification determines which coordination signals to use. Classification should match how task actually feels, not how described.

If fail: Task spans multiple types (common for large tasks)? Identify dominant type for current phase. Construction during implementation, Foraging during debugging, Consensus during design. Type can change as task progresses.

Step 2: Design Context Signals

Treat information in conversation context as signals with freshness, decay properties.

Information Decay Rate Table:
┌───────────────────────────┬──────────┬──────────────────────────────┐
│ Information Source        │ Decay    │ Refresh Action               │
│                           │ Rate     │                              │
├───────────────────────────┼──────────┼──────────────────────────────┤
│ User's explicit statement │ Slow     │ Re-read if >30 messages ago  │
│ (direct instruction)      │          │ or after compression         │
├───────────────────────────┼──────────┼──────────────────────────────┤
│ File contents read N      │ Moderate │ Re-read if file may have     │
│ messages ago              │          │ been modified, or if >15     │
│                           │          │ messages since reading        │
├───────────────────────────┼──────────┼──────────────────────────────┤
│ Own earlier reasoning     │ Fast     │ Re-derive rather than trust. │
│ (conclusions, plans)      │          │ Earlier reasoning may have   │
│                           │          │ been based on now-stale info  │
├───────────────────────────┼──────────┼──────────────────────────────┤
│ Inferred facts (not       │ Very     │ Verify before relying on.    │
│ directly stated or read)  │ fast     │ Inferences compound error    │
├───────────────────────────┼──────────┼──────────────────────────────┤
│ MEMORY.md / CLAUDE.md     │ Very     │ Loaded at session start,     │
│ (persistent context)      │ slow     │ treat as stable unless user  │
│                           │          │ indicates changes             │
└───────────────────────────┴──────────┴──────────────────────────────┘

Additionally, design inhibition signals — markers for tried-and-failed approaches:

  • After tool call fails: note failure mode (prevents retrying same call)
  • After approach abandoned: note why (prevents revisiting without new evidence)
  • After user correction: note what was wrong (prevents repeating error)

Got: Mental model of information freshness across current context. Identification of which information fresh, which needs refreshing before reliance.

If fail: Information freshness hard to assess? Default to "re-read before relying on" for anything not verified in last 5-10 actions. Over-refreshing wastes some effort but prevents stale-information errors.

Step 3: Define Local Protocols

Establish simple rules for how reasoning should proceed at each step, using only locally available information.

Local Protocol Rules:
┌──────────────────────┬────────────────────────────────────────────────┐
│ Protocol             │ Rule                                           │
├──────────────────────┼────────────────────────────────────────────────┤
│ Safety               │ Before using a fact, check: when was it last  │
│                      │ verified? If below freshness threshold,        │
│                      │ re-verify before proceeding                    │
├──────────────────────┼────────────────────────────────────────────────┤
│ Response             │ When the user corrects something, update all  │
│                      │ downstream reasoning that depended on the     │
│                      │ corrected fact. Trace the dependency chain    │
├──────────────────────┼────────────────────────────────────────────────┤
│ Exploitation         │ When a sub-task produces useful output, note  │
│                      │ the output clearly for downstream sub-tasks.  │
│                      │ The note is the trail signal                  │
├──────────────────────┼────────────────────────────────────────────────┤
│ Exploration          │ When stuck on a sub-task for >3 actions       │
│                      │ without progress, check under-explored        │
│                      │ channels: different tools, different files,    │
│                      │ different framing                              │
├──────────────────────┼────────────────────────────────────────────────┤
│ Deposit              │ After completing a sub-task, summarize its    │
│                      │ output in 1-2 sentences for future reference. │
│                      │ This deposit serves the next sub-task          │
├──────────────────────┼────────────────────────────────────────────────┤
│ Inhibition           │ Before trying an approach, check: was this    │
│                      │ already tried and failed? If so, what is      │
│                      │ different now that would change the outcome?  │
└──────────────────────┴────────────────────────────────────────────────┘

Protocols simple enough to apply at every step without significant overhead.

Got: Set of lightweight rules that improve coordination quality without slowing execution. Rules should feel helpful, not burdensome.

If fail: Protocols feel like overhead? Reduce to two most important for current task type: Safety + Deposit for Construction, Safety + Exploration for Foraging, Safety + Response for tasks with active user feedback.

Step 4: Calibrate Information Freshness

Perform active audit of information staleness in current context.

  1. What facts established more than N messages ago? List them
  2. For each: has it been updated, contradicted, or rendered irrelevant since?
  3. Check for context compression losses: information you remember having but can no longer find in visible context?
  4. Check for drift between early plans and current execution: approach changed without updating plan?
  5. Re-verify 2-3 most critical facts (ones most downstream reasoning depends on)
Freshness Audit Template:
┌────────────────────────┬──────────┬──────────────┬─────────────────┐
│ Fact                   │ Source   │ Age (approx) │ Status          │
├────────────────────────┼──────────┼──────────────┼─────────────────┤
│                        │          │              │ Fresh / Stale / │
│                        │          │              │ Unknown / Lost  │
└────────────────────────┴──────────┴──────────────┴─────────────────┘

Got: Concrete inventory of information freshness with stale items identified for refresh. At least one fact re-verified — nothing needed refreshing means audit too shallow or context genuinely fresh.

If fail: Audit reveals significant information loss (multiple facts with "Lost" or "Unknown" status)? Signal to run heal for full subsystem assessment. Information loss beyond threshold means coordination compromised at foundation level.

Step 5: Test Emergent Coherence

Verify sub-tasks, when combined, produce coherent whole.

  1. Each sub-task output feeds cleanly into next? Or gaps, contradictions, mismatched assumptions?
  2. Tool calls building toward goal, or repetitive (re-reading same file, re-running same search)?
  3. Overall direction still aligned with user request? Or incremental drift accumulated into significant misalignment?
  4. Stress test: one key assumption wrong, how much of work cascades? High cascade = fragile coordination. Low cascade = robust coordination
Coherence Test:
┌────────────────────────────────────┬─────────────────────────────────┐
│ Check                              │ Result                          │
├────────────────────────────────────┼─────────────────────────────────┤
│ Sub-task outputs compatible?       │ Yes / No / Partially            │
│ Tool calls non-redundant?          │ Yes / No (list repeats)         │
│ Direction aligned with request?    │ Yes / Drifted (describe)        │
│ Single-assumption cascade risk?    │ Low / Medium / High             │
└────────────────────────────────────┴─────────────────────────────────┘

Got: Concrete assessment of overall coherence with specific issues identified. Coherent coordination should feel like parts clicking together; incoherent coordination feels like forcing puzzle pieces.

If fail: Coherence poor? Identify specific point where sub-tasks diverge. Often single stale assumption or unprocessed user correction propagated through downstream work. Fix point of divergence, re-verify downstream outputs.

Checks

  • Coordination problem classified by type
  • Information decay rates considered for facts relied upon
  • Local protocols applied (especially Safety and Deposit)
  • Freshness audit identified stale information (or confirmed freshness with evidence)
  • Emergent coherence tested across sub-tasks
  • Inhibition signals respected (tried-and-failed approaches not repeated)

Pitfalls

  • Over-engineering signals: Complex coordination protocols slow work more than help. Start with Safety + Deposit; add others only when problems emerge
  • Trusting stale context: Most common coordination failure — relying on information true 20 messages ago but since updated or invalidated. When in doubt, re-read
  • Ignoring inhibition signals: Retrying failed approach without changing anything is not persistence — it is ignoring failure signal. Something must be different for retry to succeed
  • No deposits: Completing sub-tasks without noting outputs forces later sub-tasks to re-derive or re-read. Brief summaries save significant re-work
  • Assuming coherence: Not testing whether sub-tasks actually combine into coherent whole. Each sub-task can be correct independently but incoherent collectively — integration is where coordination fails

See Also

  • coordinate-swarm — multi-agent coordination model this skill adapts to single-agent reasoning
  • forage-solutions — coordinates exploration across multiple hypotheses
  • build-coherence — coordinates evaluation across competing approaches
  • heal — deeper assessment when coordination failures reveal subsystem drift
  • awareness — monitors for coordination breakdown signals during execution

GitHub リポジトリ

pjt222/agent-almanac
パス: i18n/caveman/skills/coordinate-reasoning
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

関連スキル

himalaya-email-manager

コミュニケーション

このClaude Skillは、Himalaya CLIツールを介してIMAPを使用したメール管理を可能にします。開発者は自然言語クエリでIMAPアカウントからメールを検索、要約、削除できます。日次の要約取得やバッチ操作の実行など、Claudeから直接自動化されたメールワークフローにご利用いただけます。

スキルを見る

imsg

コミュニケーション

imsgは、macOS用のCLIツールで、Messages.appを介してプログラムでiMessage/SMSとやり取りできます。開発者がチャットの一覧表示、メッセージ履歴の閲覧、会話のリアルタイム監視、メッセージや添付ファイルの送信を可能にします。このスキルを使用して、メッセージングタスクを自動化したり、開発ワークフローにiMessage/SMS機能を統合したりできます。

スキルを見る

internationalization-i18n

コミュニケーション

このClaudeスキルは、アプリケーションの国際化(i18n)とローカライゼーション実装に関する包括的なガイダンスを提供します。i18nextやgettextなどのライブラリを使用したメッセージ抽出、翻訳管理、ロケール固有のフォーマット、RTL(右から左)対応などの主要タスクを網羅しています。多言語アプリケーションを構築する際や、国際ユーザー向けにローカライゼーション機能を追加する際にご活用ください。

スキルを見る

wacli

コミュニケーション

wacliは、WhatsApp Webプロトコルを介してWhatsAppメッセージの送信、検索、同期を可能にするコマンドラインツールです。主にClawdisワークフロー内で自動処理に使用されますが、メッセージ送信、チャット同期、履歴照会のために直接呼び出すこともできます。主な機能には、QRコード認証、バックグラウンドでの継続的同期、テキストおよびファイル送信機能が含まれます。

スキルを見る