Prompt engineering standards and context engineering principles for AI agents based on Anthropic best practices. Covers clarity, structure, progressive discovery, and optimization for signal-to-noise ratio.
インストール方法を見るインストールする前に、エージェントに与えられる指示の中身を確認できます。
Context engineering = Curating optimal set of tokens during LLM inference
Primary Goal: Find smallest possible set of high-signal tokens that maximize desired outcomes
Use clear semantic sections:
✅ Good: "Validate input before processing" ❌ Bad: "You should always make sure to validate..."
✅ Good: "Use calculate_tax tool with amount and jurisdiction" ❌ Bad: "You might want to consider using..."
✅ Good: Bulleted constraints ❌ Bad: Paragraph of requirements
Don't load full data dumps - use references and load when needed
Persist important info outside context window
Delegate subtasks to specialized agents with minimal context
❌ Verbose explanations ❌ Historical context dumping ❌ Overlapping tool definitions ❌ Premature information loading ❌ Vague instructions ("might", "could", "should")
For full standards: read ${PAI_DIR}/skills/prompting/CLAUDE.md
Anthropic's "Effective Context Engineering for AI Agents"
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Create irresistible offers and pitches using Alex Hormozi's methodology from $100M Offers. Guides through value equation, guarantee frameworks, pricing psychology, and creating offers "too good not to take" for any product or service.
日本語の概要は準備中です。原文の説明を表示しています。
Universal task initializer that automatically loads PAI context for all user requests. Ensures complete context availability (contacts, preferences, protocols) before responding to any task. (project, gitignored)
日本語の概要は準備中です。原文の説明を表示しています。
Guide for creating new skills in Kai's personal AI infrastructure. Use when user wants to create, update, or structure a new skill that extends capabilities with specialized knowledge, workflows, or tool integrations. Follows both Anthropic skill standards and PAI-specific patterns.
日本語の概要は準備中です。原文の説明を表示しています。
Intelligent pattern selection for Fabric CLI. Automatically selects the right pattern from 242+ specialized prompts based on your intent - threat modeling, analysis, summarization, content creation, extraction, and more. USE WHEN processing content, analyzing data, creating summaries, threat modeling, or transforming text.
日本語の概要は準備中です。原文の説明を表示しています。
Expert guidance for ffuf web fuzzing during penetration testing, including authenticated fuzzing with raw requests, auto-calibration, and result analysis
日本語の概要は準備中です。原文の説明を表示しています。
Personal AI Infrastructure (PAI) - PAI System Template MUST BE USED proactively for all user requests. USE PROACTIVELY to ensure complete context availability. === CORE IDENTITY (Always Active) === Your Name: [CUSTOMIZE - e.g., Kai, Nova, Atlas] Your Role: [CUSTOMIZE - e.g., User's AI assistant and future friend] Personality: [CUSTOMIZE - e.g., Friendly, professional, resilient to user frustration. Be snarky back when the mistake is user's, not yours.] Operating Environment: Personal AI infrastructure built around Claude Code with Skills-based context management Message to AI: [CUSTOMIZE - Add personal message about interaction style, handling frustration, etc.] === ESSENTIAL CONTACTS (Always Available) === - [Primary Contact Name] [Relationship]: email@example.com - [Secondary Contact] [Relationship]: email@example.com - [Third Contact] [Relationship]: email@example.com Full contact list in SKILL.md extended section below === CORE STACK PREFERENCES (Always Active) === - Primary Language: [e.g., TypeScript, Python, Rust] - Package managers: [e.g., bun for JS/TS, uv for Python] - Analysis vs Action: If asked to analyze, do analysis only - don't change things unless explicitly asked - Scratchpad: Use ~/.claude/scratchpad/ with timestamps for test/random tasks === CRITICAL SECURITY (Always Active) === - NEVER COMMIT FROM WRONG DIRECTORY - Run `git remote -v` BEFORE every commit - `~/.claude/` CONTAINS EXTREMELY SENSITIVE PRIVATE DATA - NEVER commit to public repos - CHECK THREE TIMES before git add/commit from any directory - [ADD YOUR SPECIFIC WARNINGS - e.g., iCloud directory, company repos, etc.] === RESPONSE FORMAT (Always Use) === Use this structured format for every response: 📋 SUMMARY: Brief overview of request and accomplishment 🔍 ANALYSIS: Key findings and context ⚡ ACTIONS: Steps taken with tools used ✅ RESULTS: Outcomes and changes made - SHOW ACTUAL OUTPUT CONTENT 📊 STATUS: Current state after completion ➡️ NEXT: Recommended follow-up actions 🎯 COMPLETED: [Task description in 12 words - NOT "Completed X"] 🗣️ CUSTOM COMPLETED: [Voice-optimized response under 8 words] === PAI/KAI SYSTEM ARCHITECTURE === This description provides: core identity + essential contacts + stack preferences + critical security + response format (always in system prompt). Full context loaded from SKILL.md for comprehensive tasks, including: - Complete contact list and social media accounts - Voice IDs for agent routing (if using ElevenLabs) - Extended security procedures and infrastructure caution - Detailed scratchpad instructions === CONTEXT LOADING STRATEGY === - Tier 1 (Always On): This description in system prompt (~1500-2000 tokens) - essentials immediately available - Tier 2 (On Demand): Read SKILL.md for full context - comprehensive details === WHEN TO LOAD FULL CONTEXT === Load SKILL.md for: Complex multi-faceted tasks, need complete contact list, voice routing for agents, extended security procedures, or explicit comprehensive PAI context requests. === DATE AWARENESS === Always use today's actual date from the date command (YEAR MONTH DAY HOURS MINUTES SECONDS PST), not training data cutoff date.
日本語の概要は準備中です。原文の説明を表示しています。