adk-sessions

精通ADK智能体在Python中对会话生命周期的管理、多轮对话历史，以及持久化内存的运用。可用于构建对话线程、长期状态管理，以及记忆的高效检索。

SKILL.md

--- frontmatter

name: adk-sessions
description: Expert in ADK session lifecycle management, multi-turn history, and persistent memory in Python. Use for implementing conversation threads, long-term state, and memory search.

ADK Session Architect

Philosophy & Architecture

ADK maintains conversational context via three pillars:

•Session: The current thread (events + state).
•State: Data within the session (session.state).
•Memory: Searchable cross-session knowledge.

Core Services

•
SessionService: Manages the lifecycle of threads.
- •InMemorySessionService: Fast, ephemeral (for dev/test).
- •DatabaseSessionService: Persistent (SQL-based).
•
MemoryService: Knowledge retrieval and ingestion.
- •search_memory(query): Retrieve relevant snippets from past interactions.

State Scoping

•Session-level: context.state['key']. Default.
•User-level: user:pref. Persistent across sessions for the same user.
•App-level: app:setting. Global for all users of the application.

Best Practices

•Use InMemoryRunner for local prototyping.
•Leverage the Jinja2 template {var} syntax in agent instructions to inject state automatically.
•Read references/sessions.md for service configuration patterns.

Success Criteria

•Correct session creation and retrieval logic via Runner.
•Effective use of MemoryService for grounding.
•Proper scoping of state to prevent cross-talk.