416 skills · sorted by relevance
Challenge important design and implementation decisions with a separate reviewer that does not inherit the working session's context. Identify hidden assumptions and failure cases, then assess each finding against the artifact and its requirements.
Measure code performance and run benchmarks to investigate slow operations, regressions, or memory pressure. Establish a baseline, compare results under matching conditions, report variability, and rerun correctness tests after optimization.
Benchmark how CodeGraph affects an agent investigating a real codebase. Choose a version, language, repository, and execution mode, then compare cost, tool calls, file reads, and answer correctness with and without CodeGraph.
Evaluate an Agent Skill bundle before release by checking structure, activation conditions, output quality, scripts, safety, file integrity, and host compatibility. Receive a detailed JSON report that distinguishes local evidence checks from production readiness backed by external attestation.
Review code changes in a diff, pull request, branch, or file path to identify bugs and cleanup opportunities. Findings include file locations and concrete failure scenarios, with options to apply fixes or post comments.
Fix white flashes and blending problems between scenes in HyperFrames videos. This skill explains background setup, overlapping scene containers, final-frame holds, and timeline templates so transitions render with the intended appearance.
Review code changes through a structured check of requirements, design, correctness, security, performance, and tests. Produce specific, constructive feedback that separates required fixes from optional suggestions, and build shared review guidelines for a development team.
Simplify code once its behavior is understood. Confirm the existing behavior through tests or reproduction, reduce unnecessary complexity in small, reviewable steps, and keep verification passing while preserving behavior and necessary safety checks.
Review a pull request or code diff for bugs by reading callers, expected behavior, and nearby error handling. Receive findings tied to specific lines, ranked by severity, plus a summary of merge risks and needed tests.
Add a CUDA/C++ GPU operation to sgl-kernel using AOT, meaning compilation before use. Follow the implementation through PyTorch registration, Python API exposure, SGLang integration, correctness tests, and benchmarks that account for cache reuse.
Guides language model training with GRPO in TRL, comparing several responses and rewarding better results. Covers dataset preparation, scoring rules for correctness and format, training configuration, monitoring, and saving models for later use.
Evaluate code generation models with HumanEval, MBPP, and other benchmarks. Test generated programs, compare models across programming languages, and record pass@k scores, which estimate the chance that at least one of k solutions passes the tests.
Assess GitHub issues and pull requests by checking demonstrated user needs, existing alternatives, compatibility, and maintenance costs. Produce an evidence-based maintainer recommendation through code and document inspection, without running tests or examples.
Edit videos through conversation, from transcription and take selection to cuts, color adjustments, animated overlays, and subtitles. Confirm an editing plan, review a checked preview, and refine the result before the final render.
Reviews code changes through documented engineering and product principles associated with named practitioners. Combines two engineering perspectives and one product perspective to produce findings with severity, attribution confidence, and technical reasoning for architecture, performance, or user experience issues.
Review agent-written code changes before merge, focusing on concrete correctness defects. Uses independent subagent reviews, targeted verification, and fresh checks after substantial repairs, with a deeper evidence-based process for security, concurrency, persistence, or data-loss risks.
Review skill and agent definitions for structural issues and whether their instructions meet expected behavior. Four modes cover structure, test specifications, category criteria, and coverage reports that reveal missing specifications or checks.
Test user journeys affected by a branch’s changes in a real browser. Check correctness and the experience for intended users, fix small defects with regression tests, and commit fixes alongside a report documenting results and unresolved decisions.
Compare multiple NemoClaw pull requests addressing the same issue. It checks merge eligibility, correctness, quality, and attribution, then produces an evidence-based recommendation for merging or choosing a candidate for further repair.
Review Handsontable code changes for architecture, code quality, performance, accessibility, and test coverage. Findings include severity, confidence, and file locations, helping developers assess whether changes follow project rules and demonstrate the intended behavior.
Review code changes for unnecessary complexity, including unused flexibility, avoidable dependencies, and abstractions with little purpose. Receive concise findings that name what to remove, what can replace it, and how many lines could be cut.
Improve MNN inference on OpenCL by measuring bottlenecks and refining GPU kernels, the code that performs calculations. The skill guides baseline measurements, correctness checks, Android device benchmarks, and integration of new OpenCL features with compatibility fallbacks.
Guide MNN Metal backend development and optimization, from finding performance bottlenecks to changing GPU computation, combining operators, and adjusting execution scheduling. A task-based document index directs work toward relevant implementation rules, benchmarks, and correctness checks.
Optimize and troubleshoot MNN's Hexagon DSP backend, which runs computation on a dedicated processor. The workflow uses device measurements to identify bottlenecks, compare v79 and v81 architectures, verify correctness, and investigate crashes or memory usage.