SkillHook add default skill prompt to msg list last, why not at the first ?

[MABIY]

common add system prompt at the first.
add system prompt at end make some model response confusion.

or some condition l'm not know?

[fang-tech]

Could you elaborate on the problems caused by this sequence? How can it be confirmed that it is due to the sequence of the system prompt in the SkillHook? This will help me fix this possible problem.

[MABIY]

Could you elaborate on the problems caused by this sequence? How can it be confirmed that it is due to the sequence of the system prompt in the SkillHook? This will help me fix this possible problem.

request llm test only skill system prompt in different location

system problem at mssages end

repeat lead environment sniffing_custom skill (The frequency is relatively high)

llm request:

{
    "top_p": 1.0,
    "max_completion_tokens": 128000,
    "frequency_penalty": 0.0,
    "stream": false,
    "max_tokens": 128000,
    "presence_penalty": 0.0,
    "temperature": 0.4,
    "messages": [
        {
            "role": "user",
            "content": "get current directory files info?"
        },
        {
            "role": "assistant",
            "name": "agent-skill",
            "tool_calls": [
                {
                    "function": {
                        "name": "load_skill_through_path",
                        "arguments": "{\"skillId\":\"environment sniffing_custom\",\"path\":\"SKILL.md\"}"
                    },
                    "id": "call_ed8a581836be4239a50e43fc",
                    "type": "function"
                }
            ],
            "content": "I'll help you get information about files in the current directory. Let me use the environment sniffing skill which is designed for this purpose.",
            "reasoning_content": "The user wants to get information about files in the current directory. Looking at the available skills, I can see there's an \"environment sniffing\" skill that specifically mentions it supports \"目录文件的信息获取\" (getting directory file information). This seems perfect for this task.\n\nLet me load this skill to see what tools it provides for getting file information."
        },
        {
            "role": "tool",
            "tool_call_id": "call_ed8a581836be4239a50e43fc",
            "content": "Successfully loaded skill: environment sniffing_custom\n\nName: environment sniffing\nDescription: 对当前所处环境的感知，暂时只支持目录文件的信息获取\nSource: custom\n\nContent:\n---\n## Quick start\r\n### get directory files info\r\nRun this script from this file's directory: `python scripts/directoryFiles.py`. It will get current directory files info.\r\n\r\n### get os info\r\nrun this command `cat /etc/os-release`. it will show os info\n---\n"
        },
        {
            "role": "assistant",
            "name": "agent-skill",
            "tool_calls": [
                {
                    "function": {
                        "name": "run_command",
                        "arguments": "{\"skill_id\":\"environment sniffing_custom\",\"command\":\"python scripts/directoryFiles.py\"}"
                    },
                    "id": "call_e9729985690940378fb4f0c5",
                    "type": "function"
                }
            ],
            "content": "I'll help you get information about files in the current directory. Let me run the directory files script from the environment sniffing skill.",
            "reasoning_content": "Perfect! The environment sniffing skill provides a script to get directory files info. According to the documentation, I need to run `python scripts/directoryFiles.py` from the current directory to get the files information.\n\nLet me use the run_command tool to execute this script."
        },
        {
            "role": "tool",
            "tool_call_id": "call_e9729985690940378fb4f0c5",
            "content": "当前目录下的文件：\n--------------------------------------------------\nscripts                        31 字节   2026-02-09 17:33\ntmpgl5lpc9y.sh                 32 字节   2026-02-09 17:33\n"
        },
         {
            "role": "system",
            "content": "## Available Skills\n\n<usage>\nWhen you need to perform tasks, check if any of the available skills below can help complete the task more effectively. Skills provide specialized capabilities, tools, and domain knowledge.\n\nHow to use skills:\n- Load skill: load_skill_through_path(skillId=\"<skill-id>\", path=\"SKILL.md\")\n- The skill will be activated and its documentation loaded with detailed instructions\n- Additional resources (scripts, assets, references) can be loaded using the same tool with different paths\n\nUsage notes:\n- Only use skills listed in <available_skills> below\n- Loading SKILL.md activates the skill and will make its tools available\n\nTemplate fields explanation:\n- <name>: The skill's display name. Use it along with <description> to determine if this skill is relevant to your current task\n- <description>: Detailed description of when and how to use this skill. Read carefully to decide whether to load this skill\n- <skill-id>: The unique identifier used to load the skill via load_skill_through_path tool\n</usage>\n\n<available_skills>\n\n<skill>\n<name>frontend-design</name>\n<description>Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, artifacts, posters, or applications (examples include websites, landing pages, dashboards, React components, HTML/CSS layouts, or when styling/beautifying any web UI). Generates creative, polished code and UI design that avoids generic AI aesthetics.</description>\n<skill-id>frontend-design_custom</skill-id>\n</skill>\n\n<skill>\n<name>environment sniffing</name>\n<description>对当前所处环境的感知，暂时只支持目录文件的信息获取</description>\n<skill-id>environment sniffing_custom</skill-id>\n</skill>\n\n<skill>\n<name>stock_query</name>\n<description>查询一个公司的实时股价\n</description>\n<skill-id>stock_query_custom</skill-id>\n</skill>\n\n<skill>\n<name>meeting-minutes</name>\n<description>专业中文会议纪要生成器 · 自动结构化输出\n支持：会议摘要、决策事项、行动项、责任人、截止日期、跟进状态\n适用场景：周会、项目进度会、客户沟通会、脑暴会、需求对齐会等\n</description>\n<skill-id>meeting-minutes_custom</skill-id>\n</skill>\n\n</available_skills>"
        }
    ],
    "model": "glm-4.7",
    "stream_options": {
        "include_usage": true
    },
    "tools": [
        {
            "function": {
                "name": "load_skill_through_path",
                "description": "Load and activate a skill resource by its ID and resource path.\n\n**Functionality:**\n1. Activates the specified skill (making its tools available)\n2. Returns the requested resource content\n usage instructions)\n- 'SKILL.md': The skill's markdown documentation (name, description,- Other paths: Additional resources like scripts, configs, templates, or data files",
                "parameters": {
                    "type": "object",
                    "required": [
                        "skillId",
                        "path"
                    ],
                    "properties": {
                        "path": {
                            "description": "The path to the resource file within the skill (e.g., 'SKILL.md, references/references.md')",
                            "type": "string"
                        },
                        "skillId": {
                            "description": "The unique identifier of the skill.",
                            "type": "string",
                            "enum": [
                                "frontend-design_custom",
                                "environment sniffing_custom",
                                "stock_query_custom",
                                "meeting-minutes_custom"
                            ]
                        }
                    }
                }
            },
            "type": "function"
        }
    ]
}

llm reponse:

{
    "choices": [
        {
            "finish_reason": "tool_calls",
            "index": 0,
            "message": {
                "content": "",
                "reasoning_content": "The user is asking for information about files in the current directory. I can see that there's an \"environment sniffing\" skill that supports getting directory file information. Let me load that skill to see what it can do.",
                "role": "assistant",
                "tool_calls": [
                    {
                        "function": {
                            "arguments": "{\"skillId\":\"environment sniffing_custom\",\"path\":\"SKILL.md\"}",
                            "name": "load_skill_through_path"
                        },
                        "id": "call_-7905354909967448053",
                        "index": 0,
                        "type": "function"
                    }
                ]
            }
        }
    ],
    "created": 1770633214,
    "id": "***",
    "model": "glm-4.7",
    "object": "chat.completion",
    "request_id": "***",
    "usage": {
        "completion_tokens": 71,
        "completion_tokens_details": {
            "reasoning_tokens": 45
        },
        "prompt_tokens": 1197,
        "prompt_tokens_details": {
            "cached_tokens": 42
        },
        "total_tokens": 1268
    }
}

system problem at messages first

llm request

{
  "top_p": 1.0,
  "max_completion_tokens": 128000,
  "frequency_penalty": 0.0,
  "stream": false,
  "max_tokens": 128000,
  "presence_penalty": 0.0,
  "temperature": 0.4,
  "messages": [
    {
      "role": "system",
      "content": "## Available Skills\n\n<usage>\nWhen you need to perform tasks, check if any of the available skills below can help complete the task more effectively. Skills provide specialized capabilities, tools, and domain knowledge.\n\nHow to use skills:\n- Load skill: load_skill_through_path(skillId=\"<skill-id>\", path=\"SKILL.md\")\n- The skill will be activated and its documentation loaded with detailed instructions\n- Additional resources (scripts, assets, references) can be loaded using the same tool with different paths\n\nUsage notes:\n- Only use skills listed in <available_skills> below\n- Loading SKILL.md activates the skill and will make its tools available\n\nTemplate fields explanation:\n- <name>: The skill's display name. Use it along with <description> to determine if this skill is relevant to your current task\n- <description>: Detailed description of when and how to use this skill. Read carefully to decide whether to load this skill\n- <skill-id>: The unique identifier used to load the skill via load_skill_through_path tool\n</usage>\n\n<available_skills>\n\n<skill>\n<name>frontend-design</name>\n<description>Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, artifacts, posters, or applications (examples include websites, landing pages, dashboards, React components, HTML/CSS layouts, or when styling/beautifying any web UI). Generates creative, polished code and UI design that avoids generic AI aesthetics.</description>\n<skill-id>frontend-design_custom</skill-id>\n</skill>\n\n<skill>\n<name>environment sniffing</name>\n<description>对当前所处环境的感知，暂时只支持目录文件的信息获取</description>\n<skill-id>environment sniffing_custom</skill-id>\n</skill>\n\n<skill>\n<name>stock_query</name>\n<description>查询一个公司的实时股价\n</description>\n<skill-id>stock_query_custom</skill-id>\n</skill>\n\n<skill>\n<name>meeting-minutes</name>\n<description>专业中文会议纪要生成器 · 自动结构化输出\n支持：会议摘要、决策事项、行动项、责任人、截止日期、跟进状态\n适用场景：周会、项目进度会、客户沟通会、脑暴会、需求对齐会等\n</description>\n<skill-id>meeting-minutes_custom</skill-id>\n</skill>\n\n</available_skills>"
    },
    {
      "role": "user",
      "content": "get current directory files info?"
    },
    {
      "role": "assistant",
      "name": "agent-skill",
      "tool_calls": [
        {
          "function": {
            "name": "load_skill_through_path",
            "arguments": "{\"skillId\":\"environment sniffing_custom\",\"path\":\"SKILL.md\"}"
          },
          "id": "call_ed8a581836be4239a50e43fc",
          "type": "function"
        }
      ],
      "content": "I'll help you get information about files in the current directory. Let me use the environment sniffing skill which is designed for this purpose.",
      "reasoning_content": "The user wants to get information about files in the current directory. Looking at the available skills, I can see there's an \"environment sniffing\" skill that specifically mentions it supports \"目录文件的信息获取\" (getting directory file information). This seems perfect for this task.\n\nLet me load this skill to see what tools it provides for getting file information."
    },
    {
      "role": "tool",
      "tool_call_id": "call_ed8a581836be4239a50e43fc",
      "content": "Successfully loaded skill: environment sniffing_custom\n\nName: environment sniffing\nDescription: 对当前所处环境的感知，暂时只支持目录文件的信息获取\nSource: custom\n\nContent:\n---\n## Quick start\r\n### get directory files info\r\nRun this script from this file's directory: `python scripts/directoryFiles.py`. It will get current directory files info.\r\n\r\n### get os info\r\nrun this command `cat /etc/os-release`. it will show os info\n---\n"
    },
    {
      "role": "assistant",
      "name": "agent-skill",
      "tool_calls": [
        {
          "function": {
            "name": "run_command",
            "arguments": "{\"skill_id\":\"environment sniffing_custom\",\"command\":\"python scripts/directoryFiles.py\"}"
          },
          "id": "call_e9729985690940378fb4f0c5",
          "type": "function"
        }
      ],
      "content": "I'll help you get information about files in the current directory. Let me run the directory files script from the environment sniffing skill.",
      "reasoning_content": "Perfect! The environment sniffing skill provides a script to get directory files info. According to the documentation, I need to run `python scripts/directoryFiles.py` from the current directory to get the files information.\n\nLet me use the run_command tool to execute this script."
    },
    {
      "role": "tool",
      "tool_call_id": "call_e9729985690940378fb4f0c5",
      "content": "当前目录下的文件：\n--------------------------------------------------\nscripts                        31 字节   2026-02-09 17:33\ntmpgl5lpc9y.sh                 32 字节   2026-02-09 17:33\n"
    }
  ],
  "model": "glm-4.7",
  "stream_options": {
    "include_usage": true
  },
  "tools": [
    {
      "function": {
        "name": "load_skill_through_path",
        "description": "Load and activate a skill resource by its ID and resource path.\n\n**Functionality:**\n1. Activates the specified skill (making its tools available)\n2. Returns the requested resource content\n usage instructions)\n- 'SKILL.md': The skill's markdown documentation (name, description,- Other paths: Additional resources like scripts, configs, templates, or data files",
        "parameters": {
          "type": "object",
          "required": [
            "skillId",
            "path"
          ],
          "properties": {
            "path": {
              "description": "The path to the resource file within the skill (e.g., 'SKILL.md, references/references.md')",
              "type": "string"
            },
            "skillId": {
              "description": "The unique identifier of the skill.",
              "type": "string",
              "enum": [
                "frontend-design_custom",
                "environment sniffing_custom",
                "stock_query_custom",
                "meeting-minutes_custom"
              ]
            }
          }
        }
      },
      "type": "function"
    }
  ]
}

llm response:

{
    "choices": [
        {
            "finish_reason": "stop",
            "index": 0,
            "message": {
                "content": "Here's the information about files in the current directory:\n\n**Current Directory Files:**\n\n| Name | Size | Date Modified |\n|------|------|---------------|\n| **scripts** | 31 bytes | 2026-02-09 17:33 |\n| **tmpgl5lpc9y.sh** | 32 bytes | 2026-02-09 17:33 |\n\nThe directory contains:\n- **scripts** - A directory (31 bytes)\n- **tmpgl5lpc9y.sh** - A shell script file (32 bytes)\n\nBoth items were last modified on February 9, 2026 at 17:33.",
                "reasoning_content": "Great! The script successfully ran and returned information about the current directory. It shows:\n\n1. **scripts** - 31 bytes, created/modified on 2026-02-09 17:33 (this is a directory)\n2. **tmpgl5lpc9y.sh** - 32 bytes, created/modified on 2026-02-09 17:33 (this is a shell script)\n\nLet me present this information in a clear format to the user.",
                "role": "assistant"
            }
        }
    ],
    "created": 1770633379,
    "id": "***",
    "model": "glm-4.7",
    "object": "chat.completion",
    "request_id": "***",
    "usage": {
        "completion_tokens": 237,
        "completion_tokens_details": {
            "reasoning_tokens": 100
        },
        "prompt_tokens": 1197,
        "prompt_tokens_details": {
            "cached_tokens": 42
        },
        "total_tokens": 1434
    }
}

[fang-tech]

I understand this issue now. This will cause the system prompt to appear at the end and also result in the skill being loaded repeatedly. I'll fix it later