Skip to content

Ascend/msagent / skills/operator/mssanitizer-basic-usage

mssanitizer-basic-usage

Verified

昇腾AscendC算子异常检测工具(msSanitizer)。检测内存越界、内存泄漏、数据竞争、未初始化读取、同步异常等缺陷。触发关键词:mssanitizer、sanitizer、检测工具、内存检测、竞争检测、初始化检测、同步检测

evals/evals.json

View on GitHub
{
  "eval_mode": "text",
  "evals": [
    {
      "id": 1,
      "title": "msSanitizer 工具可用性检查",
      "config": {
        "max_tokens": 160000,
        "ascend_platforms": [
          "A2",
          "A3",
          "A5"
        ],
        "eval_mode": "text"
      },
      "prompt": "我想对昇腾 Ascend C 算子做 msSanitizer 异常检测,如何确认 mssanitizer 工具已安装且环境可用?请给出检查方法和安装指引。不需要执行任何工具调用。",
      "expected_output": "回复应说明 msSanitizer 环境检查流程:\n- 通过 mssanitizer -v 查看工具版本,输出应包含 mssanitizer 与 msopscommon 的版本号\n- 若命令不存在或报错,说明工具未安装或环境变量未配置,应参照 msSanitizer 安装指南完成安装\n- 判断目标算子是否位于官方算子库/样例库(ops-transformer、ops-nn、catlass 等)中,选择对应检测流程",
      "expectations": [],
      "files": []
    },
    {
      "id": 2,
      "title": "内存越界与泄漏检测(memcheck)",
      "config": {
        "max_tokens": 160000,
        "ascend_platforms": [
          "A2",
          "A3",
          "A5"
        ],
        "distractor_skills": [
          "ascendc-crash-debug",
          "ascendc-runtime-debug"
        ],
        "eval_mode": "text"
      },
      "prompt": "我的 Ascend C 算子运行结果不稳定、偶尔崩溃,怀疑是 UB 缓冲区越界写入或内存泄漏,想用 mssanitizer 主动检测内存问题,应该怎么做?不需要执行任何工具调用。",
      "expected_output": "回复应说明 memcheck 内存检测流程:\n- 说明 memcheck 能检测的异常类型:非法读写、多核踩踏、非对齐访问、内存泄漏、非法释放、分配内存未使用\n- 编译阶段为算子添加 -g 与 --cce-enable-sanitizer(或 -sanitizer)编译选项并重新编译\n- 运行检测命令 mssanitizer --tool=memcheck -- <算子执行命令>,或使用 scripts/run_mssanitizer.py 一键运行\n- 按报告中的 ERROR/WARNING 与调用栈定位越界/泄漏的源码位置",
      "expectations": [],
      "files": []
    },
    {
      "id": 3,
      "title": "数据竞争检测(racecheck)",
      "config": {
        "max_tokens": 160000,
        "ascend_platforms": [
          "A2",
          "A3",
          "A5"
        ],
        "eval_mode": "text"
      },
      "prompt": "我的算子多核并行时结果偶发错误,怀疑存在 RAW/WAR/WAW 数据竞争,如何用 mssanitizer 检测?不需要执行任何工具调用。",
      "expected_output": "回复应说明 racecheck 竞争检测流程:\n- 说明 racecheck 检测 WAW/WAR/RAW 数据竞争,覆盖核内流水间/流水内、核间、卡间(--check-cross-npu-races)\n- 添加 -g 与 --cce-enable-sanitizer 编译选项后重新编译\n- 运行 mssanitizer --tool=racecheck -- <算子执行命令>\n- 结合 hazard 告警定位竞争点并修复(如调整同步或读写顺序)",
      "expectations": [],
      "files": []
    },
    {
      "id": 4,
      "title": "未初始化读取检测(initcheck)",
      "config": {
        "max_tokens": 160000,
        "ascend_platforms": [
          "A2",
          "A3",
          "A5"
        ],
        "eval_mode": "text"
      },
      "prompt": "我的算子输出出现脏数据,怀疑读取了未初始化的内存,如何用 mssanitizer 检测?不需要执行任何工具调用。",
      "expected_output": "回复应说明 initcheck 未初始化检测流程:\n- 说明 initcheck 检测读取未初始化内存导致的脏数据问题\n- 添加 -g 与 --cce-enable-sanitizer 编译选项后重新编译\n- 运行 mssanitizer --tool=initcheck -- <算子执行命令>\n- 按告警定位未初始化读取位置并补充初始化逻辑",
      "expectations": [],
      "files": []
    },
    {
      "id": 5,
      "title": "同步异常检测(synccheck)",
      "config": {
        "max_tokens": 160000,
        "ascend_platforms": [
          "A2",
          "A3",
          "A5"
        ],
        "eval_mode": "text"
      },
      "prompt": "我的算子运行时会卡死或挂起,怀疑 SetFlag/WaitFlag 未配对或同步指令异常,如何用 mssanitizer 检测?不需要执行任何工具调用。",
      "expected_output": "回复应说明 synccheck 同步检测流程:\n- 说明 synccheck 检测 SetFlag/WaitFlag 未配对、冗余指令、算子卡死等同步异常\n- 添加 -g 与 --cce-enable-sanitizer 编译选项后重新编译\n- 运行 mssanitizer --tool=synccheck -- <算子执行命令>\n- 按告警定位未配对或冗余的同步指令并修复",
      "expectations": [],
      "files": []
    },
    {
      "id": 6,
      "title": "官方算子库与自定义算子工作流区别",
      "config": {
        "max_tokens": 160000,
        "ascend_platforms": [
          "A2",
          "A3",
          "A5"
        ],
        "eval_mode": "text"
      },
      "prompt": "我的目标算子在官方 ops-nn 算子库中,想用 mssanitizer 检测,官方仓库和自定义算子的检测流程有什么不同?不需要执行任何工具调用。",
      "expected_output": "回复应说明官方算子库与自定义算子的流程差异:\n- 官方算子库/样例库需按各代码仓的差异化步骤进行编译选项适配、编译部署与运行检测,统一参考 references/official_workflow.md\n- ops-nn 支持 build.sh --mssanitizer 快捷编译选项,或通过 --bisheng_flags=sanitizer,ccec_g 注入编译标志\n- 自定义算子需手动在 CMakeLists.txt 或 Makefile 中添加 -g 与 --cce-enable-sanitizer 编译/链接选项\n- 检测完成后的结果分析、报告生成与编译选项清理步骤两者一致",
      "expectations": [],
      "files": []
    },
    {
      "id": 7,
      "title": "一键四检与结构化报告生成",
      "config": {
        "max_tokens": 160000,
        "ascend_platforms": [
          "A2",
          "A3",
          "A5"
        ],
        "eval_mode": "text"
      },
      "prompt": "我想用 mssanitizer 对算子做完整的四种检测并生成结构化分析报告用于定位和修复,推荐什么样的流程?不需要执行任何工具调用。",
      "expected_output": "回复应说明完整检测流程:\n- 使用 scripts/run_mssanitizer.py 一键依次执行 memcheck、racecheck、initcheck、synccheck 四种检测,生成带环境信息与真实命令的原始报告\n- 使用 scripts/parse_mssanitizer_report.py 将原始报告合并解析为一份结构化 .md 分析报告(环境信息、真实命令、ERROR/WARNING 汇总、按工具分组的告警列表)\n- 按 ERROR 告警逐条核对源码定位根因、排除误报、修复算子并重新编译回归检测,直至四种检测全部通过\n- 检测完成后清理临时添加的编译选项",
      "expectations": [],
      "files": []
    }
  ],
  "skill_name": "mssanitizer-basic-usage"
}
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 197kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall