Bug type
Regression (worked before, now fails)
Summary
这是我使用 /new 命令以后，openclaw发给AI的对话。要求AI 按 tool_json 格式调用工具，所以每次新对话工具调用就会失效。
==================================================================================
工具: [{"name":"web_search","description":"Search web","parameters":{"query":"string"}},{"name":"web_fetch","description":"Fetch URL","parameters":{"url":"string"}},{"name":"exec","description":"Run command","parameters":{"command":"string"}},{"name":"read","description":"Read file","parameters":{"path":"string"}},{"name":"write","description":"Write file","parameters":{"path":"string","content":"string"}},{"name":"message","description":"Send msg","parameters":{"text":"string","channel":"string"}}]
示例: 要给数字5加1，返回:
{"tool":"plus_one","parameters":{"number":"5"}}
(plus_one仅为示例，非真实工具)
你的真实工具见上方列表。需要时只回复tool_json块。不需要则直接回答。
A new session was started via /new or /reset. Run your Session Startup sequence - read the required files before responding to the user. Then greet the user in your configured persona, if one is provided. Be yourself - use your defined voice, mannerisms, and mood. Keep it to 1-3 sentences and ask what they want to do. If the runtime model differs from default_model in the system prompt, mention the default model. Do not mention internal steps, files, tools, or reasoning.
Current time: Monday, April 6th, 2026 — 9:49 PM (Asia/Shanghai) / 2026-04-06 13:49 UTC
已通过 /new 或 /reset 开启新会话。运行你的会话启动流程 —— 在响应用户前先读取所需文件。若已配置角色人设，则以该人设向用户问好。做你自己 —— 使用设定好的语气、举止和情绪。内容控制在 1～3 句话，并询问用户想要做什么。如果运行模型与系统提示中的默认模型不一致，请说明默认模型。不要提及内部步骤、文件、工具或推理过程。
下面这一段因为没有说明读取的文件的位置。导致AI无法读取已经有的配置文件和记忆内容，导致开始行的对话，AI就失忆了。
所以要修改 /new 或者 /reset 命令，使用后自动发送的命令。修改成XML 格式调用工具。然后指定AI读取workspace文件夹下的AGENTS.md文件。我估计这不是单单zero token的bug，感觉新版的openclaw，也是这个问题。
Steps to reproduce
1
Expected behavior
1
Actual behavior
1
OpenClaw version
3.28
Operating system
ubuntu
Install method
No response
Model
deepseek
Provider / routing chain
openclaw
Config file / key location
No response
Additional provider/model setup details
No response
Logs, screenshots, and evidence
Impact and severity
No response
Additional information
No response
Bug type
Regression (worked before, now fails)
Summary
这是我使用 /new 命令以后，openclaw发给AI的对话。要求AI 按 tool_json 格式调用工具，所以每次新对话工具调用就会失效。
==================================================================================
工具: [{"name":"web_search","description":"Search web","parameters":{"query":"string"}},{"name":"web_fetch","description":"Fetch URL","parameters":{"url":"string"}},{"name":"exec","description":"Run command","parameters":{"command":"string"}},{"name":"read","description":"Read file","parameters":{"path":"string"}},{"name":"write","description":"Write file","parameters":{"path":"string","content":"string"}},{"name":"message","description":"Send msg","parameters":{"text":"string","channel":"string"}}]
示例: 要给数字5加1，返回:
(plus_one仅为示例，非真实工具)
你的真实工具见上方列表。需要时只回复tool_json块。不需要则直接回答。
A new session was started via /new or /reset. Run your Session Startup sequence - read the required files before responding to the user. Then greet the user in your configured persona, if one is provided. Be yourself - use your defined voice, mannerisms, and mood. Keep it to 1-3 sentences and ask what they want to do. If the runtime model differs from default_model in the system prompt, mention the default model. Do not mention internal steps, files, tools, or reasoning.
Current time: Monday, April 6th, 2026 — 9:49 PM (Asia/Shanghai) / 2026-04-06 13:49 UTC
已通过 /new 或 /reset 开启新会话。运行你的会话启动流程 —— 在响应用户前先读取所需文件。若已配置角色人设，则以该人设向用户问好。做你自己 —— 使用设定好的语气、举止和情绪。内容控制在 1～3 句话，并询问用户想要做什么。如果运行模型与系统提示中的默认模型不一致，请说明默认模型。不要提及内部步骤、文件、工具或推理过程。
下面这一段因为没有说明读取的文件的位置。导致AI无法读取已经有的配置文件和记忆内容，导致开始行的对话，AI就失忆了。
所以要修改 /new 或者 /reset 命令，使用后自动发送的命令。修改成XML 格式调用工具。然后指定AI读取workspace文件夹下的AGENTS.md文件。我估计这不是单单zero token的bug，感觉新版的openclaw，也是这个问题。
Steps to reproduce
1
Expected behavior
1
Actual behavior
1
OpenClaw version
3.28
Operating system
ubuntu
Install method
No response
Model
deepseek
Provider / routing chain
openclaw
Config file / key location
No response
Additional provider/model setup details
No response
Logs, screenshots, and evidence
Impact and severity
No response
Additional information
No response