スキル一覧に戻る

vishnu-bhaga

pjt222
更新日 Yesterday
2 閲覧
17
2
17
GitHubで表示
デザインaidesign

について

ヴィシュヌ・バーガスキルは、安定した作業状態を維持し、スコープクリープやコンテキストドリフトから保護するため一貫性を強制します。このスキルは検証済みの知識を固定し、特に破壊的な変更後や長時間のセッションにおいて継続性を保証します。機能するシステムを修正前に安定化させるため、または重要な決定が失われるのを防ぐためにご活用ください。

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/vishnu-bhaga

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

Vishnu Bhaga

Preserve, sustain what is working — anchor verified knowledge, maintain consistency under perturbation, protect functional patterns from unnecessary change.

When Use

  • Working approach at risk of being disrupted by scope creep or premature optimization
  • Context drift threatening to overwrite verified knowledge with stale assumptions
  • Multiple parallel concerns creating pressure to change things that should remain stable
  • After shiva-bhaga dissolution — what survives needs active protection during reconstruction
  • Long session risks losing earlier verified decisions through context compression
  • Before making changes to system currently functioning correct

Inputs

  • Required: Current working state or verified knowledge to preserve (available implicit)
  • Optional: Specific threat to stability (e.g., "scope creep," "context compression approaching")
  • Optional: MEMORY.md and project files for grounding (via Read)

Steps

Step 1: Inventory What Works

Before protecting anything, identify what is currently functional and verified.

Preservation Inventory:
+---------------------+---------------------------+------------------------+
| Category            | Verification Method       | Anchoring Action       |
+---------------------+---------------------------+------------------------+
| Verified Facts      | Confirmed via tool use    | Record source and      |
|                     | (file reads, test runs,   | timestamp; do not      |
|                     | API responses)            | re-derive              |
+---------------------+---------------------------+------------------------+
| Working Code        | Tests pass, behavior      | Do not refactor unless |
|                     | confirmed, user approved  | explicitly requested   |
+---------------------+---------------------------+------------------------+
| User Requirements   | Explicitly stated by      | Quote directly; do not |
|                     | the user in this session  | paraphrase or infer    |
+---------------------+---------------------------+------------------------+
| Agreed Decisions    | Decisions made and        | Reference the decision |
|                     | confirmed during this     | point; do not revisit  |
|                     | session                   | without new evidence   |
+---------------------+---------------------------+------------------------+
| Environmental State | File paths, configs,      | Verify before assuming |
|                     | tool availability         | unchanged              |
+---------------------+---------------------------+------------------------+
  1. For each category, list specific items currently verified and working
  2. Note verification method — how do you know this is true?
  3. Items without verification not preserved — are assumptions (and may need shiva-bhaga)

Got: Concrete inventory of verified, working elements with their evidence base.

If err: Inventory sparse — little is verified? Itself valuable information. Run heal to re-ground before attempting to preserve unverified assumptions.

Step 2: Identify Perturbation Sources

Name forces threatening stable state.

  1. Scope creep: Task expanding beyond what was agreed?
  2. Context drift: Earlier facts being overwritten by more recent (possibly incorrect) reasoning?
  3. Optimization pressure: Urge to improve something working adequate?
  4. External changes: Environment changed (files modified, tools unavailable)?
  5. Compression risk: Conversation approaching context limits where early decisions may be lost?

For each source, assess: real threat or anticipated one?

Got: Named perturbation sources with assessed severity (active threat vs. anticipated risk).

If err: No perturbation sources apparent? Preservation may not be needed — consider whether brahma-bhaga (creation) or continued execution more appropriate.

Step 3: Anchor Stable State

Apply specific techniques to protect what works from identified threats.

  1. Memory anchoring: For critical facts at risk of context drift, re-state explicitly:
    • "Established fact: [X], verified by [method] at [point in conversation]"
    • Persistent memory available? Write durable facts to MEMORY.md
  2. Scope boundary enforcement: For scope creep, re-state agreed scope:
    • "Agreed scope: [original request]. Current work is within/outside this boundary."
  3. Change resistance: For working code under optimization pressure:
    • "This component working and tested. No changes unless user requests them."
  4. State snapshot: For compression risk, create mental checkpoint:
    • Summarize: what done, what remains, what key decisions made
  5. Environmental verification: For external changes, re-check before proceeding:
    • Re-read critical files rather than relying on earlier reads

Got: Each identified threat has specific anchoring response. Stable state explicitly protected.

If err: Anchoring feels excessive — protecting everything equal? Prioritize. What is one thing that must not change? Protect that first.

Step 4: Sustain Through Action

Preservation not passive — needs ongoing attention during subsequent work.

  1. Before each action, check: "Does this threaten anything in preservation inventory?"
  2. Yes? Find alternative approach achieving goal without disturbing stable state
  3. Disturbance unavoidable? Acknowledge explicit and update inventory
  4. Periodically re-verify preserved items — especially after complex operations
  5. When task completes, confirm preserved items remain intact

Got: Working state survives current task intact. Changes made only where needed. Did not disrupt functioning components.

If err: Preserved item inadvertently changed? Assess damage immediate. Change broke something? Revert. Change neutral? Update inventory. Do not leave inventory stale.

Check

  • Working state inventoried with verification evidence
  • Perturbation sources identified and assessed
  • Anchoring actions applied to each real threat
  • Scope boundaries maintained throughout task
  • Preserved items re-verified after completion

Pitfalls

  • Preserve assumptions as facts: Only verified knowledge deserves protection. Unverified assumptions dressed as facts create false stability
  • Over-preservation: Protecting everything equally prevents necessary change. Preservation must be selective — protect what works, release what does not
  • Passive preservation: Assuming things will stay stable without active verification. Context drift constant. Preservation needs ongoing attention
  • Resistance to legitimate change: Using preservation as excuse to avoid necessary modifications. User requests change to working component? Overrides preservation
  • Stale inventory: Failing to update preservation inventory as new information arrives. Inventory must reflect current reality, not state at creation time

See Also

  • shiva-bhaga — destruction precedes preservation; what survives dissolution is what Vishnu sustains
  • brahma-bhaga — creation builds on preserved foundation; new patterns emerge from stable ground
  • heal — subsystem assessment reveals what is genuinely functional vs. superficially stable
  • observe — sustained neutral observation detects drift before it threatens stability
  • awareness — situational awareness (Cooper color codes) maps direct to perturbation detection

GitHub リポジトリ

pjt222/agent-almanac
パス: i18n/caveman/skills/vishnu-bhaga
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

関連スキル

executing-plans

デザイン

executing-plansスキルは、完全な実装計画があり、それを管理されたバッチでレビューチェックポイントを設けながら実行する場合に使用します。このスキルは計画を読み込んで批判的にレビューした後、小さなバッチ(デフォルトは3タスク)でタスクを実行し、各バッチの間に進捗状況を報告してアーキテクトのレビューを受けます。これにより、品質管理チェックポイントが組み込まれた体系的な実装が保証されます。

スキルを見る

requesting-code-review

デザイン

このスキルは、コードレビュアーサブエージェントを起動し、処理を進める前に要件に対してコード変更を分析します。タスク完了後、主要な機能の実装後、またはmainブランチへのマージ前などに使用すべきです。このレビューは、現在の実装と元の計画を比較することで、問題を早期に発見するのに役立ちます。

スキルを見る

connect-mcp-server

デザイン

このスキルは、開発者がHTTP、stdio、またはSSEトランスポートを使用してMCPサーバーをClaude Codeに接続するための包括的なガイドを提供します。GitHub、Notion、カスタムAPIなどの外部サービスを統合するためのインストール、設定、認証、セキュリティについて解説しています。MCP統合のセットアップ、外部ツールの設定、またはClaudeのModel Context Protocolを扱う際にご利用ください。

スキルを見る

web-cli-teleport

デザイン

このスキルは、タスク分析に基づいて開発者がClaude Code WebとCLIインターフェースの選択を支援し、これらの環境間でのシームレスなセッションテレポーテーションを可能にします。Web、CLI、モバイル環境を切り替える際のセッション状態とコンテキストを管理することで、ワークフローを最適化します。様々な段階で異なるツールを必要とする複雑なプロジェクトにご活用ください。

スキルを見る