性能
性能分析与优化
rocprof
spcl
What the AMD device trace from profile returns for a hip or OpenMP-offload submission, which fields come back null, and what each refusal means.
nsys
spcl
What the NVIDIA device trace from profile returns for a cuda submission (tool nsys), what it cannot answer, and what each refusal means.
solver
spcl
"Solver kernel rules: sequential dependence vs. free loop, why reordering changes the answer not the speed, why fewer iterations is wrong. Use for Newton, Krylov, multigrid, or ODE solver kernels."
profiling
spcl
CPU profiling -- where the time went (perf) and what the machine did there (PAPI counters, per-thread CPI and imbalance).
code-polisher
k1lgor
Use this when the user asks to refactor, clean up, optimize, or improve code quality.
release-metrics
lodekeeper
Evaluate Lodestar release candidate readiness by comparing beta/RC metrics against stable. Use when deploying a new RC to beta nodes, reviewing metrics before cutting a release, or assessing whether a release candidate has regressions. Covers health checks, performance comparison, memory/resource analysis, validator effectiveness, networking quality, and PeerDAS-specific metrics. Requires Grafana/Prometheus access to the Lodestar monitoring stack.
opt-reports
spcl
Get the compiler's own optimization report for your submission from the judge, and tell a legality refusal from a cost-model one.
component-patterns
armanzeroeight
Implement React component patterns including composition, custom hooks, render props, HOCs, and compound components. Use when building reusable React components, implementing design patterns, or refactoring component architecture.
resume-updater
Aznatkoiny
Conversational skill that interviews the user to capture professional experience and build/update their career-profile.json. Use when the user says "update my resume", "add my recent experience", "refresh my profile", "capture my work history", "build my resume", "add a new job to my resume", or wants to create or modify their professional profile data. When to Use: - User wants to add new experience, skills, projects, or achievements - User wants to create their career profile from scratch - User wants to update existing profile entries - User mentions they changed jobs, got promoted, or completed a project When NOT to Use: - User wants to generate/format a resume (use /resume-generator command) - User wants career advice (use career-director agent) - User wants to search for jobs (use job-search agent)
cpp-reinforcement-learning
Aznatkoiny
C++ Reinforcement Learning best practices using libtorch (PyTorch C++ frontend) and modern C++17/20. Use when: - Implementing RL algorithms in C++ for performance-critical applications - Building production RL systems with libtorch - Creating replay buffers and experience storage - Optimizing RL training with GPU acceleration - Deploying RL models with ONNX Runtime
performance-scaling
bejranonda
Cross-model performance optimization and scaling configurations for autonomous agents
kanban-orchestrator
magnusthovsland
Decomposition playbook + anti-temptation rules for an orchestrator profile routing work through Kanban. The "don't do the work yourself" rule and the basic lifecycle are auto-injected into every kanban worker's system prompt; this skill is the deeper playbook when you're specifically playing the orchestrator role.
databases
mamba-mental
Work with MongoDB (document database, BSON documents, aggregation pipelines, Atlas cloud) and PostgreSQL (relational database, SQL queries, psql CLI, pgAdmin). Use when designing database schemas, writing queries and aggregations, optimizing indexes for performance, performing database migrations, configuring replication and sharding, implementing backup and restore strategies, managing database users and permissions, analyzing query performance, or administering production databases.
swiftui-performance-audit
dagba
Audit and improve SwiftUI runtime performance from code review and architecture. Use for requests to diagnose slow rendering, janky scrolling, high CPU/memory usage, excessive view updates, or layout thrash in SwiftUI apps, and to provide guidance for user-run Instruments profiling when code review alone is insufficient.
take-the-brief
krishanraja
"Exclusion gate first: do not invoke when Krish says not to interview him and supplies or identifies an existing brief, even if he also says 'take this off my plate', 'own it', or 'use your judgment'; route onward to strategy/execution. Otherwise use this intent-briefing gateway when Krish explicitly asks to be interviewed before execution, asks for an AI employee, or transfers new high-stakes/ambiguous work whose plausible interpretations materially differ. Ask up to five high-signal questions one at a time, then produce an amnesia-proof intent summary and short plan. An explicit interview request takes precedence. Do not use for clear single-step, urgent, routine, already-briefed work, customer interviews, or durable taste-standard elicitation; use ctrl-intake for the last."
elicitation
tasteray
'Psychological profiling through natural conversation using narrative identity
recommendations
tasteray
'TasteRay API integration for personalized recommendations across verticals
challenge-me
milistu
Direct technical advisor mode for counting/estimation and architecture/design
agent-memory-skills
kimasplund
Self-improving agent architecture using ChromaDB for continuous learning, self-evaluation, and improvement storage. Agents maintain separate memory collections for learned patterns, performance metrics, and self-assessments without modifying their static .md configuration.
session-close
mseok
"End-of-session checklist. Checks for uncommitted changes, offers to update focus, project docs, context, and session log. Replaces the old stop reminder flow."
seed-workspace
camshaft
How to build, run, test, and gate the Cadenza seed toolchain via cargo xtask. Read this whenever working in the Rust workspace under implementation/seed/crates (cadenza-syntax, rcdzc, cdz-run, cdz-runtime) or in xtask/ — compiling or running a .cdz/.sexp program, running the corpus behavior gate, checking library health, formatting, round-tripping the syntax surfaces, or inspecting a binary AST.
integrated-reasoning-v2
kimasplund
Enhanced meta-orchestration for selecting and combining reasoning patterns. Now includes 9 methodologies (ToT, BoT, SRC, HE, AR, DR, AT, RTR, NDF) with weighted multi-dimensional selection, feedback loops, uncertainty propagation, and validated confidence aggregation. Use when facing complex problems requiring optimal reasoning strategy selection.
hypothesis-elimination
kimasplund
Systematic elimination reasoning for diagnosis and debugging. Use when you need to identify THE cause among many possibilities through evidence-based elimination. Ideal for production incidents, bug diagnosis, root cause analysis, and differential diagnosis. Unlike BoT (which explores), HE eliminates. Example: "Server is slow" → Generate 10 hypotheses, design discriminating tests, eliminate 9, confirm 1.
gh-profile
DimitriGilbert
Expert GitHub profile customization assistant that helps create stunning, professional GitHub profile READMEs. Use this skill when users want to customize their GitHub profile, create profile READMEs, add stats cards, animations, badges, or automate profile updates with GitHub Actions. Covers 2025 trends including minimalist designs, themed profiles, dynamic content, animations, visitor tracking, AI-powered generation, and GitHub achievements.