「perf」の検索結果

对 profiling 数据（kernel_details.csv / trace_view.json / step_trace_time.csv）做统一分析： 5 层递进管道（analyze_trace.py）+ baseline/优化后 kernel diff（compare_traces.py），输出 瓶颈诊断、算子执行序、可融合候选与 P0-P2 方向建议。输入统一来自 profiling-collect。 只要用户有 profiling 产出并问"为什么慢/瓶颈在哪/优化前后差多少/往哪个方向"，都用本技能；采集走 profiling-collect，方案选档走 dit-perf-opt，实现/并行/基准走 pattern-dev/dit-parallel-opt/benchmark-dev——本技能只做诊断与方向。 由 model-auto-optimization 的 S1（融合分析）/S3（并行）阶段调用，亦由 dev-workflow 的分析阶段指引加载。 融合机会候选只做**识别**；候选的**边界与收益判定**（能不能融成一个单元、融多大、值不值）交 `fusion-scope-analyze`（本技能的辅助技能）——本技能不判融合边界。

日本語の概要は準備中です。原文の説明を表示しています。

MindIE-SD 核心算子（FA/BSA/GMM/MM）性能基准工具链。使用：模型优化（model-auto-optimization 的 S1/S4 选型）中用 mindie_bench 对单算子做实现级实测，按 dtype/量化档/稀疏度/形态对比 选出最优配置（产物：选型证据；稀疏度-性能曲线供 S4 稀疏度选型）；开发：benchmarks/ 工具链扩展与新算子接入测试（供 operator-dev / pattern-dev 调用）。 当用户需要对比算子实现选型、新增或修改 benchmark 代码、排查 benchmark 数据异常、 给基准加算子/指标时使用；即使用户只说"benchmark 数据不对""给基准加个算子" "对比下这几个 FA 实现哪个快"也应触发；特性级方案选档请走 dit-perf-opt （本 skill 只做实现级实测）。 由 model-auto-optimization 的 S1/S4 选型场景与 operator-dev 的算子接入验证场景调用， 亦由 dev-workflow 在基准开发时指引加载。

日本語の概要は準備中です。原文の説明を表示しています。

量化格式契约与位级对齐：从设备字节反推量化器的**编码公式、舍入模式、scale 粒度、退化块规则**， 并据此重实现/融合量化器（MXFP8 e8m0、int8 perblock、fp8 perchannel 等）， 用「同进程同数据 + 独立参考实现 + 中点与退化输入全覆盖」做到**逐字节精确**。 当自研量化 kernel 与框架算子**对不上**、需要逆向某个量化器的数值契约、需要判断 「量化残差算不算可接受」、或要在融合内核里复现 aclnn 量化语义时使用此 skill。 即使用户只说「量化结果和参考不一致」「这个 scale 怎么算出来的」「融合量化器精度对不上」 而未提"契约"，也应触发。**注意**：单纯选量化档位/配置（该不该开量化、开哪一档）走 dit-perf-opt；算子级性能调优与 DSL 选型走 operator-dev； 并行作用域导致的静默失效（DiT 侧「改了并行但不报错也没生效」的判定）走 dit-parallel-opt。

日本語の概要は準備中です。原文の説明を表示しています。

共享 Ascend NPU 上的**性能验收标准**：判定"某优化到底有没有效、有多少效"，并规定 **只有验收态结果才能写入总览表**。分双态：**探索态**（平时特性分析：不展开、 允许少步/单次，产物必须标 `[探索]`、不得进总览表）vs **验收态**（入库：同窗 ≥N 复现、按 window-ab-protocol 口径出数）。当用户问"这个优化有多少收益 / 是否有效 / 要不要采纳 / 为什么变慢了""A-B 一下""同窗对比""测 3 次取平均""把数据对齐到同一窗口" "数字对不对""是不是我眼花了"，或需要在多卡共享机器上给出性能结论、刷新优化总览报表、 复核别人给的加速比时使用；即使用户只说"帮我确认下这个改动值不值"也应触发。 核心口径：同窗 A/B 是唯一权威（跨窗口绝对值不可比）、A/B/A 漂移校正、窗口内自带基线、 三次证明（HTTP 200 + 字节数 + 日志行）、产物 md5 与关键张量 dump 双证据、噪声地板内 不下结论、profiling 三张表统一到同一窗口；并覆盖 md5 门禁的盲区（忠实度对照）、 共享机器纪律与门控回退纪律。

日本語の概要は準備中です。原文の説明を表示しています。

Deep analysis of competitor patent claims: element decomposition, scope interpretation, equivalents prediction, prosecution history estoppel analysis. Automatically converts PDF/DOCX/DOC patent files to structured Markdown via patent-pdf-to-md skill before analysis. Invoke when user asks to analyze/parse competitor patent claims, decompose claim elements, assess claim scope, or perform 权利要求解析/要素分解/保护范围解读. 竞品专利权利要求深度解析：要素分解、保护范围解读、等同范围预判、审查历史限制分析。自动调用patent-pdf-to-md技能将PDF/DOCX/DOC专利文件转换为结构化Markdown后再进行分析。当用户要求解析竞品专利权利要求、要素分解、保护范围时触发。

日本語の概要は準備中です。原文の説明を表示しています。

Drafts a recordable Abstract of Judgment to create a judgment lien on a debtor's real property. Extracts party names, monetary components, and judgment details from case documents, then applies jurisdiction-specific formatting and certification requirements. Use post-judgment in commercial litigation when enforcing monetary awards, perfecting judgment liens, or preparing lien filings with the county recorder.

日本語の概要は準備中です。原文の説明を表示しています。

AI compliance analysis for EU AI Act, ISO 42001, NIST AI RMF, GDPR, OECD, financial services regulations (SEC, FCA, FINRA, DORA, MiFID II), and other frameworks. Use when asked to generate a compliance checklist for an AI tool or use case, determine if a risk assessment is required, score an AI tool's risk level, identify where an AI tool or use case could run afoul of regulatory requirements, perform a gap analysis, recommend remediation steps, assess a vendor, draft an acceptable use policy, map training requirements, or review jurisdiction-specific AI rules. Triggers on phrases like "compliance checklist", "risk assessment", "risk score", "does this comply", "EU AI Act", "ISO 42001", "NIST AI RMF", "AI governance", "gap analysis", "vendor assessment", "acceptable use policy", "MNPI", "AI training requirements", or any request to evaluate an AI tool or use case for regulatory compliance or risk.

日本語の概要は準備中です。原文の説明を表示しています。

Comprehensive reference for inter-process communication mechanisms -- every way processes can talk on a computer. Covers sockets (TCP, UDP, Unix domain), WebSockets, SSE, pipes (named/anonymous), shared memory (mmap, shm_open), message queues, signals, D-Bus, XPC (macOS), gRPC, REST, stdin/stdout, file-based coordination, and clipboard. Performance benchmarks, platform availability, and code examples for each. Special focus on which IPC works best for AI agent coordination. Activate on: "IPC", "inter-process communication", "process communication", "how do I talk between processes", "Unix socket vs TCP", "shared memory", "named pipe", "WebSocket vs SSE", "gRPC vs REST", "agent coordination IPC", "XPC service", "message passing", "stdout pipe", "D-Bus", "mmap". NOT for: distributed systems design (use distributed-systems), network protocol design (use networking), message queue infrastructure like Kafka (use data-pipeline-engineer).

日本語の概要は準備中です。原文の説明を表示しています。

GOMS cognitive modeling methodology for analyzing human-computer interaction task performance

日本語の概要は準備中です。原文の説明を表示しています。

Use when CPU usage is high, memory grows unboundedly, goroutines leak, mutex contention shows in traces, or escape analysis suggests excess heap allocation. Triggers: net/http/pprof endpoint exposure, go tool pprof analysis, flamegraph generation, allocs vs inuse_space heap profiles, runtime/trace event timeline, mutex profiling, block profiling, goroutine dumps, GC pressure measurement. NOT for non-Go languages, distributed tracing (use OpenTelemetry skill), or production telemetry pipelines.

日本語の概要は準備中です。原文の説明を表示しています。

Diagnose and eliminate the GraphQL N+1 resolver explosion using Facebook DataLoader's per-request batching, plus the second-order defenses (persisted queries, depth limits, cost analysis). Use when a GraphQL query that returns N items in a list triggers N+1 database round trips, when designing nested resolvers that load associations, when a list query times out under load, or when you need to budget query cost before resolvers run. NOT for REST-style endpoint design (use api-versioning-strategy), generic database query optimization (use database-query-optimizer), or schema design (use graphql-server-architect).

日本語の概要は準備中です。原文の説明を表示しています。

Analytical methods for decomposing expert performance into cognitive and physical task components

日本語の概要は準備中です。原文の説明を表示しています。

Use when running analytical SQL over Parquet/CSV/JSON without a warehouse, replacing pandas for data wrangling, joining S3 data in-place, building local data marts, or embedding OLAP into an app. Triggers: read_parquet/read_csv setup, partitioned dataset queries, hive partitioning, glob patterns for S3, COPY TO export, attach Postgres/MySQL, UDFs in Python/R, MotherDuck cloud sync, columnar performance vs row stores. NOT for OLTP workloads (concurrent writes), distributed analytics at petabyte scale (use Spark/Trino), or vector search (use pgvector/Lance).

日本語の概要は準備中です。原文の説明を表示しています。

Central catalog of available skills with metadata, capabilities, and performance history. Provides skill discovery and lookup services. Activate on 'skill registry', 'list skills', 'skill catalog', 'available skills', 'skill metadata'. NOT for matching skills to tasks (use dag-semantic-matcher) or ranking (use dag-capability-ranker).

日本語の概要は準備中です。原文の説明を表示しています。

Matches natural language task descriptions to appropriate skills using semantic similarity, ranks candidates by fit and performance history, and maintains the skill catalog. Use when assigning skills to DAG nodes, searching for the right skill for a task, ranking competing skills, or browsing the skill catalog. Activate on "find skill", "match skill", "which skill", "skill for this task", "skill catalog", "rank skills", "best skill". NOT for executing DAGs (use dag-runtime), creating skills (use skill-architect), or grading skills (use skill-grader).

日本語の概要は準備中です。原文の説明を表示しています。

Warehouse cost reduction, auto-scaling, query optimization, and lifecycle policies for data infrastructure. Activate on: data cost, warehouse credits, cost reduction, auto-scaling, lifecycle policy, cold storage, cost monitoring, resource optimization. NOT for: query performance tuning (use data-warehouse-optimizer), batch job optimization (use batch-processing-optimizer).

日本語の概要は準備中です。原文の説明を表示しています。

Snowflake, BigQuery, clustering, partitioning, and materialized views for warehouse performance. Activate on: Snowflake, BigQuery, Redshift, query optimization, clustering, partitioning, materialized view, warehouse cost, query profile. NOT for: dbt model structure (use dbt-analytics-engineer), data modeling (use dimensional-modeler).

日本語の概要は準備中です。原文の説明を表示しています。

Apply hierarchical decomposition and structural exploitation strategies from DAG theory to complex system design, problem-solving, and abstraction challenges

日本語の概要は準備中です。原文の説明を表示しています。

Validates DAG structures, performs topological sorting, detects cycles, and resolves dependency conflicts. Uses Kahn's algorithm for optimal execution ordering. Activate on 'resolve dependencies', 'topological sort', 'cycle detection', 'dependency order', 'validate dag'. NOT for building DAGs (use dag-graph-builder) or scheduling execution (use dag-task-scheduler).

日本語の概要は準備中です。原文の説明を表示しています。

Ranks skill matches by fit, performance history, and contextual relevance. Applies multi-factor scoring including success rate, resource usage, and task alignment. Activate on 'rank skills', 'best skill for', 'skill ranking', 'compare skills', 'optimal skill'. NOT for semantic matching (use dag-semantic-matcher) or skill catalog (use dag-skill-registry).

日本語の概要は準備中です。原文の説明を表示しています。

Builds, validates, schedules, and dynamically modifies DAG execution graphs. Decomposes problems into nodes with dependencies, performs topological sorting, detects cycles, resolves conflicts, and schedules wave-based parallel execution. Use when designing a DAG structure, validating dependencies, planning execution order, or modifying a DAG at runtime. Activate on "build DAG", "plan workflow", "DAG dependencies", "topological sort", "schedule execution", "modify DAG", "replan". NOT for executing DAGs (use dag-runtime), validating outputs (use dag-quality), or matching skills to nodes (use dag-skills-matcher).

日本語の概要は準備中です。原文の説明を表示しています。

Redis caching patterns, cache-aside, write-through, TTL strategies, and invalidation. Activate on: caching, Redis, cache invalidation, cache-aside, write-through, TTL, CDN cache, stale-while-revalidate. NOT for: CDN/reverse proxy setup (use api-gateway-reverse-proxy-expert), database query optimization (use data-warehouse-optimizer).

日本語の概要は準備中です。原文の説明を表示しています。

Implement AI chatbot analytics and conversation monitoring. Use when adding conversation metrics, tracking AI usage, measuring user engagement with chat, or building conversation dashboards. Activates for AI analytics, token tracking, conversation categorization, and chat performance.

日本語の概要は準備中です。原文の説明を表示しています。

CI/CD caching optimizer for dependency caching, Docker layer caching, and build speed improvements. Activate on: CI cache, build speed, dependency caching, Docker layer cache, turbo remote cache, GitHub Actions cache, pnpm store cache. NOT for: CI/CD pipeline creation (use github-actions-pipeline-builder), deployment strategy (use blue-green-deployment-orchestrator), Docker image building (use docker-multi-stage-optimizer).

日本語の概要は準備中です。原文の説明を表示しています。