feat(prompt):guide llm read session history

[aiguozhi123456]

Summary

Enable the LLM to locate and read the full session history file on demand, mitigating context loss caused by prompt compression in multi-turn conversations.

Motivation

In long-running sessions, early messages are often compressed or dropped to fit within the context window. Once discarded, those details are irrecoverable from the LLM's perspective, leading to:

• Repeated user prompts ("I already told you...").
• Broken continuity across turns.
• Hallucinated reconstructions of prior context.

By surfacing the exact session file path in the system prompt, we give the LLM a way to pull the original, uncompressed history when it needs to recall something that no longer fits in the active context.

Changes

• context.py: Add chat_id parameter to build_system_prompt() and _get_identity(), thread it into the identity template.
• identity.md: Add one line — Session file: {{ workspace_path }}/sessions/{{ channel }}_{{ chat_id }}.jsonl — with a short grep usage hint.
• Tests: 3 new cases — exact path with channel+chat_id, None fallback (sessions/_.jsonl), and full build_messages integration.

Prompt Caching Impact

None. channel + chat_id are constant within a session, so the system prompt stays stable across turns.

[aiguozhi123456]

我记得runtime context有辨认session所需信息的，当然也可以再改改，指出有辨认信息的事实。
补充:这下应该没问题了。

[Re-bin]

Hi，感谢PR，可以把description补充一下不

[aiguozhi123456]

Hi，感谢PR，可以把description补充一下不

@Re-bin 收到，已经补充了。
BTW, 请问为什么要硬编码
You are nanobot, a helpful AI assistant.
可能不利于角色扮演。
更新: 看到了，已经修了。34e8f97

[aiguozhi123456]

1.我对于session中内容的理解有点问题
2.针对的问题已在#3304 中更合理地修复了。
遂关闭此pr。