Bug Description
When using certain models, the iac-aliyun skill is very unlikely to be automatically loaded via the skill tool when the user inputs a complex prompt involving Alibaba Cloud Terraform template generation (including VPC, VSwitch, ECS, security groups, ENI, etc.).
The iac-aliyun skill is configured with user_invocable: false and relies on the model to autonomously decide whether to load it based on the when_to_use description: "当用户涉及云资源创建、模板生成、模板解释、部署等 IaC 相关操作时" (when the user is involved in cloud resource creation, template generation, template explanation, deployment, or other IaC-related operations). The affected models appear unable to correctly identify user intent and trigger this skill.
Affected models: qwen3.7-max, kimi-k2.6, deepseek v4
Steps to Reproduce
- Configure the LLM Provider to use one of the affected models listed above
- Run
iac-code
- Enter the following prompt:
生成terraform模版，在阿里云上，region为cn-beijing.
* 创建一个vpc，cidr为172.16.0.0/16.
* 每个vpc内有三个vswitch，vswitch都在cn-beijing-k。
1. vswitchC，cidr是172.16.1.0/24。
2. vswitchPH, cidr是172.16.2.0/24。这个vswitch的eni绑定安全组，连通策略为组内隔离，允许源是172.16.2.0/24目的端口是4789的udp报文进入，允许源是172.16.2.0/24目的端口是179的tcp报文。
3. vswitchSG，cidr是172.16.3.0/24。这个vswitch的eni绑定安全组，连通策略为组内隔离，允许源是172.16.3.0/24目的端口是4789的udp报文进入，允许源是172.16.2.0/24目的端口是179的tcp报文。
* 创建四个ecs：
1. ecs1~ecs4的主网卡都在vswitchC中，ecs1的hostname是ecs10，ip是172.16.1.10和；ecs2的hostname是ecs11，ip是172.16.1.11。ecs3的hostname是vxlan-gateway-PH，ip是172.16.1.6，ecs4的hostname是vxlan-gateway-SG，ip是172.16.1.7。
2. ecs1还有两个辅助网卡，辅助网卡1在vswitchPH中，ip是172.16.2.10；辅助网卡2在vswitchSG，ip是172.16.3.10。
3. ecs2还有两个辅助网卡，辅助网卡1在vswitchPH中，ip是172.16.2.11；辅助网卡2在vswitchSG，ip是172.16.3.11。
4. ecs3还有一个辅助网卡，辅助网卡在vswitchPH, ip是172.16.2.6。
5. ecs4还有一个辅助网卡，辅助网卡在vswitchSG中，ip是172.16.3.7。
* ecs1的初始化脚本（init脚本），通过"自定义数据"传入：
0. 等待eth1和eth2网卡被挂载，且为up
1. 配置vxlan接口vxlan0，没有remote不绑定dev，ip地址为10.0.1.10，up
* ecs2的初始化脚本（init脚本），通过"自定义数据"传入：
0. 等待eth1和eth2网卡被挂载，且为up
1. 配置vxlan接口vxlan0，没有remote不绑定dev，ip地址为10.0.1.11，mac地址为12:00:01:0:1:0B，up
* ecs3的初始化脚本（init脚本），通过"自定义数据"传入：
0. 等待eth1，且为up
* ecs4的初始化脚本（init脚本），通过"自定义数据"传入：
0. 等待eth1，且为up
- Observe whether the model invokes the
skill tool to load the iac-aliyun skill
Expected Behavior
The model should recognize that the user's intent involves Alibaba Cloud resource creation and Terraform template generation, and automatically invoke the iac-aliyun skill via the skill tool. It should then follow the skill's template generation workflow to produce the Terraform template.
Actual Behavior
When using the affected models, they are very unlikely to invoke the skill tool to load the iac-aliyun skill. Instead, they directly generate Terraform code, skipping critical steps defined in the skill such as parameterization rules, template validation, and the deployment workflow.
Operating System
macOS
Python Version
3.12.7
iac-code Version
0.3.0
LLM Provider
Alibaba Cloud (DashScope / Qwen)
IaC Type
Terraform
Bug Description
When using certain models, the
iac-aliyunskill is very unlikely to be automatically loaded via theskilltool when the user inputs a complex prompt involving Alibaba Cloud Terraform template generation (including VPC, VSwitch, ECS, security groups, ENI, etc.).The
iac-aliyunskill is configured withuser_invocable: falseand relies on the model to autonomously decide whether to load it based on thewhen_to_usedescription: "当用户涉及云资源创建、模板生成、模板解释、部署等 IaC 相关操作时" (when the user is involved in cloud resource creation, template generation, template explanation, deployment, or other IaC-related operations). The affected models appear unable to correctly identify user intent and trigger this skill.Affected models:
qwen3.7-max,kimi-k2.6,deepseek v4Steps to Reproduce
iac-codeskilltool to load theiac-aliyunskillExpected Behavior
The model should recognize that the user's intent involves Alibaba Cloud resource creation and Terraform template generation, and automatically invoke the
iac-aliyunskill via theskilltool. It should then follow the skill's template generation workflow to produce the Terraform template.Actual Behavior
When using the affected models, they are very unlikely to invoke the
skilltool to load theiac-aliyunskill. Instead, they directly generate Terraform code, skipping critical steps defined in the skill such as parameterization rules, template validation, and the deployment workflow.Operating System
macOS
Python Version
3.12.7
iac-code Version
0.3.0
LLM Provider
Alibaba Cloud (DashScope / Qwen)
IaC Type
Terraform