audit-trails-for-agents
Comprehensive guide to implementing audit trails and logging for AI agents including tracing, observability, compliance, and debugging
Comprehensive guide to implementing audit trails and logging for AI agents including tracing, observability, compliance, and debugging
Systematic AI extraction prompt design expert with Socratic methodology
Self-evolving skill that enables Canifi to create, install, and manage new skills autonomously
Multi-agent architectures with LangGraph. Use when building systems with multiple collaborating agents, implementing supervisor or swarm patterns, creating hierarchical agent workflows, or designing agent handoff mechanisms. Covers langgraph-supervisor, langgraph-swarm, agent-as-tool patterns, and context engineering for multi-agent systems.
Enables Claude to send messages, manage spaces, and handle Google Chat communications
A simple example skill that demonstrates Claude Code skill structure
This skill should be used when designing context management, implementing tiered fidelity, reducing token waste, applying Four Laws patterns, creating "NOT PASSED" sections, optimizing agent context, or debugging context-related issues. Provides SOTA patterns for context-efficient multi-agent systems achieving 60-80% token reduction.
Expert in building scalable ML systems, from data pipelines and model training to production deployment and monitoring.
Build production RAG (Retrieval-Augmented Generation) pipelines with LangChain and Qdrant. Use when users ask to implement semantic search, document Q&A, build RAG systems, create vector search, implement HyDE, CRAG, or Agentic RAG patterns.
Access raw conversation history from Claude Code session storage for analysis
Full context reload. Use after /clear, after compaction, or at session start. Loads conversation history, indexes docs, maps code, scans configs, checks git. Gets you immediately helpful.
PDF-RAG API reference. REQUIRED after any failed curl/jq to localhost:8000 (404, null, jq error). Also use when uncertain about endpoint path or response shape.
生成AIが人間に対して行う作業開始時オンボーディング。AIが"説明係"ではなく"進行役"として、探索→仮説→検証→要約→未確定の明示を回し、人間が最小スキーマ(因果・境界・不変条件・壊れ方・観測)を短時間で再構築できる状態に導く。 トリガー条件: - 新しいタスクやコード変更に着手する前(「この機能を修正して」「このバグを直して」) - 未知のコードベースを理解する必要がある時 - 「オンボーディングして」「作業開始の準備をして」「コードを理解したい」 - 複雑なタスクを始める前の文脈理解が必要な時 - 「作戦ブリーフを作成して」「安全に始められるようにして」
Search and retrieve context from previous Ralph iteration transcripts. Use when you need detailed context about what happened in a previous iteration beyond what's in progress.txt. Triggers on: read transcript, previous iteration, what happened in, search transcripts.
Use when coordinating multiple AI agents for parallel work. Triggers: "multi-agent", "parallel agents", "agent coordination", "QuantumDAG", "conflict-free". NOT for: Single-agent operations or basic jj commands.
Complete RAG (Retrieval-Augmented Generation) pipeline implementation with document ingestion, vector storage, semantic search, and response generation. Supports FastAPI backends with OpenAI and Qdrant. LangChain-free architecture.
Manage Ollama models for minions. Pull new models, list available ones, switch presets, check disk usage. Use when setting up or changing models.
実装開始時のコンテキストセットアップ。関連ナレッジを検索し、効率的な実装開始を支援。新規実装タスク開始時に自動トリガーまたは明示的に呼び出し可能。