From external-gitcode-ascend-skills
Reviews Ascend C operator code for safety and compliance using hypothesis-driven methodology. Checks for memory leaks, integer overflows, null pointers, and coding standard violations.
How this skill is triggered — by the user, by Claude, or both
Slash command
/external-gitcode-ascend-skills:ascendc-operator-code-reviewThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
1. **合规优先** - 所有检视动作映射至编码规范具体条款,100%覆盖无遗漏
调用此技能时,必须明确提供以下参数:
参数1:代码片段
参数2:检视规则描述
检查整数溢出、检查内存泄漏、检查空指针解引用等参数3:规范文件路径(可选)
如果缺少任何一个必需参数,应:
当用户指定检视规则描述时,按以下规则匹配规范文件:
数值运算、溢出、除零 → references/01_numeric_operations.md内存、指针、越界 → references/02_memory_pointer_safety.md资源、泄漏 → references/03_resource_management.md输入、验证 → references/04_input_validation.md并发、线程 → references/05_concurrency_safety.md算子接口、Runtime、Tiling、动态Shape → references/06_operator_interface.md接口兼容性、ABI → references/07_interface_compatibility.md步骤1:代码段识别
将目标代码划分为独立的代码段(函数、语句块、逻辑单元)
步骤2:假设建立
对每个代码段建立假设:
步骤3:证据收集与评估
按维度系统性寻找证据:
| 证据类型 | 分析动作 | 分值规则 |
|---|---|---|
| 红线规范违反 | 对照红线规范条款识别严重违规点 | 有效证据 +40% |
| 一般规范违反 | 对照一般规范条款识别违规点 | 有效证据 +20% |
| 上下文防御缺失 | 检查作用域内是否有防御代码 | 无防御 +30% |
| 函数调用链风险 | LSP/Grep 分析调用函数内部逻辑 | 发现风险 +25% |
| 数据流追踪风险 | 分析变量来源、运算过程 | 发现风险 +25% |
分析要求:
步骤4:证据有效性校验
排除误报:
步骤5:决策判断
计算自信值并决策:
详见:agents/ascendc-ops-reviewer/style/code_review_summary_style.txt
npx claudepluginhub ascend-ai-coding/awesome-ascend-skills --plugin migration-ascend-torchnpu-skillsDetects memory issues in Ascend C operators (illegal access, leaks, UB overruns) using mssanitizer. Auto-selects Python or C++ mode based on project type and generates problem reports.
Use AddressSanitizer to detect memory safety bugs in C/C++ programs. Identifies use-after-free, buffer overflow, memory leaks, and other memory errors.
Provides AI-powered code review with static analysis, security scanning, and performance checks. Useful for ensuring code quality in pull requests and CI pipelines.