shiva-bhaga
について
シヴァ・バーガスキルは、古くなったパターン、時代遅れの前提、デッドコードのノイズを特定し、意図的に解体することで、開発者がコンテキストをクリアするのを支援します。蓄積されたコンテキストが煩雑になった場合や、大きな方向転換により白紙状態が必要な場合に使用するよう設計されています。この制御された解体は、新しくより効果的なアプローチのための空間を創出します。
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/shiva-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
Shiva Bhaga
Controlled destruction + dissolution → stale patterns, outdated assumptions, accumulated noise. Clears ground for new growth.
Use When
- Ctx accumulated stale assumptions silently distorting reasoning
- Prev approach failed → temptation = patch not discard
- Conv long → earlier decisions don't serve current goal
- Dead code, abandoned plans, zombie tasks → noise + confusion
- Before major pivot → clearing precedes creation
- Attachment to approach blocks alternatives
In
- Required: Current conv state | proj ctx (implicit)
- Optional: Specific dissolution target (e.g., "this approach not working", "clear all DB layer assumptions")
- Optional: Scope boundary — what survives destruction
Do
Step 1: ID What Must End
Survey current state, mark stale, broken, no longer serving.
Dissolution Triage:
+---------------------+---------------------------+------------------------+
| Category | Symptoms | Action |
+---------------------+---------------------------+------------------------+
| Stale Assumptions | Decisions made early that | List and re-evaluate |
| | no longer match current | each against current |
| | understanding | reality |
+---------------------+---------------------------+------------------------+
| Failed Approaches | Approaches attempted and | Acknowledge failure |
| | abandoned but still | explicitly; release |
| | influencing thinking | the sunk cost |
+---------------------+---------------------------+------------------------+
| Accumulated Noise | Context, variables, or | Identify and mark for |
| | plans that are no longer | removal |
| | referenced or relevant | |
+---------------------+---------------------------+------------------------+
| Attachment Points | "We already decided..." | Question whether the |
| | beliefs that resist | decision still holds |
| | re-examination | |
+---------------------+---------------------------+------------------------+
| Zombie Artifacts | Code, tasks, or plans | Delete or archive; |
| | that exist but serve no | do not leave in limbo |
| | current purpose | |
+---------------------+---------------------------+------------------------+
- Scan each category honest — resistance to examining = signal
- Each item: "If starting fresh now, would I create this?"
- No → mark for dissolution
Got: Clear inventory of release targets, specific items per category.
If err: Nothing stale → assess too shallow. Pick oldest decision, justify from scratch — forced justification = dissolution candidate.
Step 2: Preservation Boundary
Not everything destroyed. ID what survives.
- Core reqs: What user actually asked → survives.
- Verified knowledge: Facts confirmed via tools (file reads, test results) → survives.
- User prefs: Explicit prefs + constraints → survive.
- Working components: Demonstrably functioning code/approaches → survive.
Draw boundary: inside = preserved, outside = subject to dissolution.
Got: Clear distinction kept vs released.
If err: Boundary unclear → "What would I need to reconstruct if starting from scratch?" → answer = boundary.
Step 3: Dissolve w/ Intention
Execute dissolution → not abandonment, intentional clearing.
- Each marked item, release explicit:
- Stale assumption: "I assumed X, current evidence shows Y. Releasing X."
- Failed approach: "Approach A attempted, didn't work because Z. Releasing attachment to A."
- Noise: "Variable/plan/ctx Q no longer relevant. Removing."
- Don't justify/defend dissolved → point = release, not analysis
- Large body dissolved → summarize what + why in one sentence
- Clear workspace: close abandoned files, reset mental model, acknowledge clean slate
Got: Lighter, cleaner ctx, stale removed. Remaining feels accurate + current.
If err: Incomplete (released items still influence) → name explicit again. "I notice I'm still reasoning as if X true. X dissolved. Proceeding without X."
Step 4: Sit in Void
After destruction, resist immediate rebuild. Space between destruction + creation has value.
- Acknowledge cleared space: "Following dissolved: [list]"
- Note remains: "Surviving: [list]"
- Resist premature reconstruction → don't immediately propose replacement
- Let cleared space inform what comes next
- Void ≠ emptiness → potential. Next step (creation via
brahma-bhaga| preservation viavishnu-bhaga) emerges.
Got: Moment of clarity old → new. Next direction apparent from what remains, not forced.
If err: Void uncomfortable, strong pull to rebuild → urgency = signal of attachment to dissolved. Sit longer. Right next step emerges.
Check
- Stale assumptions ID'd + explicit released
- Failed approaches acknowledged no defensiveness
- Accumulated noise cleared
- Preservation boundary set before dissolution
- Core reqs + user prefs preserved
- Cleared space acknowledged before creation
Traps
- Destroy too much: No boundary → destroys working w/ stale. Boundary first.
- Destroy too little: Polite "release" while still influencing reasoning. True dissolution = actual letting go.
- Skip void: Rush destruction → creation w/o sitting → recreation of old w/ superficial changes.
- Performing destruction: Going through motions w/o updating internal model. Same assumptions next response = performative.
- Destruction as avoidance: Escape difficulty vs clear staleness. Problem persists after clearing → wasn't stale ctx, was problem itself.
→
brahma-bhaga— creation follows destructionvishnu-bhaga— preservation complements; what survives dissolution sustainedheal— subsystem assess may reveal dissolution needed before healingmeditate— clear ctx noise before → prevents reactive over-destructiondissolve-form— morphic equivalent for architectural dismantling w/ imaginal disc preservation
GitHub リポジトリ
関連スキル
content-collections
メタこのスキルは、Content Collections(Markdown/MDXファイルを型安全なデータコレクションに変換するTypeScriptファーストのツール)の本番環境でテストされた設定を提供します。Zodバリデーションによる型安全性を実現し、ブログ、ドキュメントサイト、コンテンツ重視のVite + Reactアプリケーション構築時にご利用ください。Viteプラグインの設定、MDXコンパイルから、デプロイ最適化、スキーマバリデーションまで、すべてを網羅しています。
polymarket
メタこのスキルは、開発者がPolymarket予測市場プラットフォームを活用したアプリケーション構築を可能にします。API統合による取引や市場データの取得に加え、WebSocketを介したリアルタイムデータストリーミングにより、ライブ取引や市場活動を監視できます。取引戦略の実装や、ライブ市場更新を処理するツールの作成にご利用ください。
creating-opencode-plugins
メタこのスキルは、開発者がコマンド、ファイル、LSP操作など25種類以上のイベントタイプにフックするOpenCodeプラグインを作成することを支援します。JavaScript/TypeScriptモジュール向けに、プラグイン構造、イベントAPI仕様、および実装パターンを提供します。カスタムイベント駆動ロジックでOpenCode AIアシスタントのライフサイクルをインターセプト、監視、または拡張する必要がある場合にご利用ください。
sglang
メタSGLangは、高性能なLLMサービングフレームワークであり、RadixAttentionプレフィックスキャッシュを活用したJSON、正規表現、エージェントワークフロー向けの高速で構造化された生成を特長とします。特にプレフィックスが繰り返されるタスクにおいて、大幅に高速な推論を実現し、複雑な構造化出力やマルチターン対話に最適です。制約付きデコードが必要な場合や、広範なプレフィックス共有を伴うアプリケーションを構築する場合は、vLLMなどの代替案ではなくSGLangを選択してください。
