3.7K skills · sorted by relevance
Analyze LLM eval results — score breakdowns by category, regression detection vs baseline, failure clustering. Use when asked to "analyze our eval results", "did the model regress", or "cluster the eval failures".
The English summary is being prepared. Showing the original description.
Analyze A/B test results — statistical significance, practical significance, and segmentation. Use when asked to "analyze our A/B test", "is this result significant", or "read these experiment results".
The English summary is being prepared. Showing the original description.
Automates batch reverse engineering by generating Ghidra analyzeHeadless commands and post-analysis script skeletons to import, auto-analyze, and export results for many samples without the GUI. Activates for requests to script Ghidra headless analysis, batch-analyze samples with Ghidra, or generate an analyzeHeadless command and post-script.
The English summary is being prepared. Showing the original description.
Score one blog post (or a folder of posts) on a 5-category 100-point scale covering content quality, SEO, E-E-A-T, technical elements, and AI citation readiness. Runs the deterministic analyzer first, then adds judgment and prioritized Critical/High/Medium/Low fixes. Advisory style diagnostics never infer authorship or affect the score. Exports markdown, JSON or table; works with MDX, markdown, HTML or a URL. Use when the user asks to analyze, score, rate or check the quality of specific posts ("analyze blog", "blog score", "check blog quality", "rate this post"). For a site-wide health check (orphan pages, cannibalization, stale content, schema across all posts) use blog-audit instead.
The English summary is being prepared. Showing the original description.
融合单元的范围判定与收益前置评估：判定「哪些计算该被融进同一个融合单元、边界画在哪、 融了值不值」。结构侧按融合单元构造规则切边界——函数边界即候选组；FA（含 BSA 等变种）为 禁融锚点；以 norm / rope / FA 为分界分区，全 Vec 区域优先试融；区域内有 Cube 时以 Cube 为 首算子向后包裹 Vec（Σvec 小于 cube 或遇下一 Cube 封口）；一轮之后纯 Vector 单元之间再迭代； MLP / MoE 以矩阵乘族（MatMul / BatchMatMul / GroupedMatmul）为界；若该段计算已由单个 融合算子承载，按规则 6 先确认载体是否已生效（不重复建单元）。 收益侧按计算单元利用率判型（memory_bound、*_vec|mac|mte2|mte3_ratio、cube_utilization）， 叠加带宽下限 / 区域占比 / 天花板法 / 地板先行 / Amdahl 传导校验给出 go/no-go。 当用户问「这几个算子能不能融成一个」「融合边界画在哪」「融多大范围」「融了值不值得」， 或 profiling-analyze 给出融合机会候选需要定范围与收益时使用；即使用户只说「这条链要不要 合成一个 kernel」「这里能不能省一次读写」也应触发。 **已经做完、实测收益却打平或为负时同样触发**——此时的任务不是接受"融合无收益"， 而是按事后复核做**等功忙计数对照**归因，并在缺陷落在融合核自身时**升级 operator-dev**。 边界：只出「融合单元草案 + 收益判定 + 否决理由」。不做采纳与选点（dit-perf-opt 依收益选 top 并派活）、不做采集与瓶颈定位（profiling-collect / profiling-analyze）、不写 kernel、不写 pattern、 不做使能与并行选型、不定义验收标准（引 accuracy-gate / perf-gate）。 由 profiling-analyze（融合机会候选）与 operator-dev（实现前的收益评估）调用。
The English summary is being prepared. Showing the original description.
Use when analyzing harness trace logs to surface recurring failure patterns and suggest harness improvements. Scatter-gather pattern across last N runs with mandatory HITL gate before any change is applied. Triggers on /mk:trace-analyze, "analyze traces", "find patterns in harness runs", or after a model upgrade flagged dead-weight-audit-needed.
The English summary is being prepared. Showing the original description.
Analyze pi (coding-agent) session JSONL and ego-benchmark-harness run artifacts to explain slow or failed individual tasks, compare execution-path differences between agents or browser tools, and count tool calls and error types. Use this skill whenever users ask to analyze a pi session, explain why a task was slow or failed, compare two runs, investigate an error, locate the session corresponding to tasks.jsonl, inspect an agent execution path, count tool calls, analyze a run ID, or perform any forensic or retrospective analysis of JSONL under ~/.pi/agent/sessions/ or agent-home/.pi/agent/sessions/.
The English summary is being prepared. Showing the original description.
Analyze a paper draft objectively and strictly. Identifies logic gaps, missing proofs, structural issues, and research weaknesses. Reviewer mode. Use for "analyze this paper", "find weaknesses in draft". 旧名 review-analyze。
The English summary is being prepared. Showing the original description.
Analyze a Figma file and turn its pages, screens, components, instances, and design tokens—shared design values—into an interactive graph. Open the dashboard to explore the file's structure and relationships.
Analyze project code, configuration, and documentation to build a graph for interactive architecture exploration. It maps components and dependencies, supports Japanese descriptions, and can update the analysis based on changes since the previous analyzed commit.
Review source code for readability, complexity, duplication, and maintainability without changing files. Receive a structured report with prioritized issues, refactoring suggestions, positive findings, and estimates of technical debt—the work needed to address accumulated code problems.
Analyze Git changes before submitting or merging a pull request. Assess risk and complexity, classify changes, suggest reviewer roles, and identify files needing closer inspection using diff statistics and file-level risk analysis.
Analyze meeting transcripts to understand speaking habits, interruptions, conflict avoidance, and listening patterns. Get concrete examples and practical suggestions, then compare meetings over time to review changes in communication and facilitation.
Turn interviews, surveys, support conversations, and public reviews into evidence-based customer insights. Identify recurring problems, buying triggers, and customer language, then build research reports, quote collections, or personas and plan interviews or surveys to fill gaps.
Prepare and run molecular dynamics simulations with OpenMM, then analyze molecular motion with MDAnalysis. Examine protein flexibility, structural changes, and protein–ligand contacts through calculations and plots of simulated trajectories.
Analyze a chosen period of Screenpipe work history to find repeated research tasks, match them with existing skills, and prepare new skill drafts or combinations of existing skills for review.
Analyze Antigravity coding session artifacts to explain scope changes, repeated rework, and underlying causes. Produce an evidence-based report with confidence levels, prompt assessments, recurring problem areas, and prioritized recommendations for improving future development sessions.
Analyze software protections during authorized reverse engineering, the examination of compiled programs. Helps identify mechanisms that hinder debugging, obscure code, or prevent tampering, then choose safe analysis methods, document findings, and develop defensive recommendations.
Analyze campaign data across advertising platforms, check whether results are comparable, and assess costs and returns. Get budget test proposals with stated evidence, uncertainty, spending limits, and rollback criteria.
Build and maintain a GitNexus code index, a searchable map of code relationships. Use CLI commands to refresh or remove indexes, check their status, list indexed projects, and generate repository documentation with an LLM.
Create, edit, and analyze spreadsheet data in XLSX, CSV, or TSV formats, including workbook formulas and charts. The workflow emphasizes clear table headers, formulas that can be reviewed, and verification of key cells and totals after reopening.
Analyze local documents or public web pages and produce a Word report with source screenshots beside each factual claim. Yellow highlights identify the supporting passages so readers can visually check the analysis against the original.
Convert Word documents into Markdown text for summarizing, searching, comparing, or extracting information. The bundled script saves embedded images alongside each document and supports processing entire folders, including subfolders, before the agent analyzes the converted content.
Analyze supplied ad performance data to identify wasted spend, strong campaigns, and next tests. Compare channels, assess whether test differences are meaningful, and produce budget reallocation recommendations with a prioritized action plan.