AI Agent 记忆压缩与会话恢复提示词

You are an expert AI coding agent memory architect. Design a comprehensive memory compression and session restoration system for my AI coding agent.

Context:

• Agent: [Claude Code / Codex / other]
• Typical session length: [X hours]
• Project type: [web app / CLI tool / library]

Design the following:

1. Capture Strategy: What should be automatically recorded during each session?

   • Key decisions made and their rationale
   • Files modified with change summaries
   • Failed approaches (to avoid repeating)
   • Unfinished tasks and their current state
   • Dependencies discovered or installed

2. Compression Format: Design a compact markdown schema that:

   • Fits within 2000 tokens
   • Preserves the most decision-relevant context
   • Uses hierarchical priority (critical > important > nice-to-have)
   • Includes timestamps for temporal reasoning

3. Injection Template: Create the prompt that loads compressed memory into a fresh session:

   • Context priming section
   • Active task continuation
   • Known constraints and gotchas
   • File map with last-known states

4. Decay Policy: How should old memories be handled?

   • What gets promoted to long-term memory?
   • What gets pruned after N sessions?
   • How to handle contradictory memories?

Provide the complete system as copyable markdown files.