Claude/OpenCode skills with eval-driven development.
Skills that enforce a specific way of working — a multi-step procedure or discipline the model follows once the skill loads.
| Skill | Summary | Baseline | With Skill | Δ | Last Run |
|---|---|---|---|---|---|
| brainstorm | Interview-driven design through relentless questioning before implementation | 67% | 100% | +33% | 2026-06-30 |
| code-reviewer | Four-reviewer parallel code review pipeline with fix cycle and aggregated findings | 100% | 100% | +0% | 2026-06-30 |
| git-commit | Conventional commit authoring: grouping changes by intent and writing commit messages | 76% | 96% | +20% | 2026-07-23 |
| git-pr | Pull request creation: sizing, title conventions, and description writing | 7% | 53% | +47% | 2026-07-23 |
| knowledge-base | Maintaining and writing to a persistent Obsidian-style cross-session knowledge graph | 47% | 100% | +53% | 2026-07-23 |
| logging | Production logging discipline: structured logs, happy-path coverage, correlation IDs | 100% | 92% | -8% | 2026-07-23 |
| orchestrator | Execute a planned epic end-to-end by dispatching coders and reviewers in autonomous waves | — | — | — | — |
| planner | Decompose a well-understood effort into a typed, critic-reviewed task graph ready for execution | — | — | — | — |
| pr-reviewer | Peer PR review pipeline producing tiered, courteous draft comments for GitHub | 40% | 100% | +60% | 2026-07-23 |
| prd-to-stories | Decomposing behavioral PRDs into deliverable stories with specific acceptance criteria | 88% | 94% | +6% | 2026-07-23 |
| prd-writing | Co-authoring behavioral product requirements documents through structured interview | 60% | 87% | +27% | 2026-07-23 |
| skill-creator | Creating, iterating, and benchmarking skills through eval-driven refinement | 33% | 27% | -7% | 2026-07-23 |
| test-driven-development | Red-green-refactor TDD workflow: writing tests before implementation | 53% | 82% | +28% | 2026-07-23 |
| wiki-hygiene | Scan the knowledge base for stale notes, orphans, and frontmatter inconsistencies | — | — | — | — |
Skills that supply frameworks and judgment for architectural or creative decisions.
| Skill | Summary | Baseline | With Skill | Δ | Last Run |
|---|---|---|---|---|---|
| agent-architecture | Framework for deciding when and how to build LLM-driven agent systems | 51% | 85% | +35% | 2026-07-23 |
| agent-evaluation | Designing evaluation systems, rubrics, and metrics for autonomous agents | 60% | 73% | +13% | 2026-07-23 |
| agent-memory | Cross-session memory design for personalized agents: storage, consolidation, retrieval | 50% | 93% | +43% | 2026-07-23 |
| agent-post-training | Decision framework for when and how to fine-tune models for agent tasks | 51% | 89% | +38% | 2026-07-23 |
| agent-self-evolution | Designing agents that accumulate experience and improve across sessions without retraining | 70% | 95% | +25% | 2026-07-23 |
| agent-tool-design | Designing tool interfaces that LLMs can reliably discover, select, and invoke | 89% | 100% | +11% | 2026-07-23 |
| api-design | REST and gRPC API conventions: resource modeling, naming, versioning, error handling | 78% | 94% | +16% | 2026-09-09 |
| application-architecture | Layering, domain modeling, and data access patterns for application-level structure | 25% | 100% | +75% | 2026-06-30 |
| coding-agent-design | Building coding agents: toolset design, security model, workflow, and failure handling | 53% | 100% | +47% | 2026-07-23 |
| concurrency-design | Choosing thread topology, concurrency units, and inter-component communication models | 67% | 93% | +27% | 2026-07-22 |
| context-engineering | Structuring LLM context windows for cache efficiency and instruction persistence | 59% | 88% | +29% | 2026-07-23 |
| game-audio | Audio design principles for games: music, sound effects, and adaptive systems | 73% | 100% | +27% | 2026-07-22 |
| game-design | Frameworks for analyzing mechanics, game feel, player motivation, and systemic design | 75% | 100% | +25% | 2026-06-30 |
| game-narrative | Applied interactive storytelling: deduction systems, NPC agency, non-linear discovery | 100% | 100% | +0% | 2026-06-30 |
| game-patterns | Engine-agnostic game programming patterns: ECS, state machines, event systems, pooling | 75% | 100% | +25% | 2026-06-30 |
| game-performance | Systematic GPU/CPU/memory profiling and optimization methodology for games | 100% | 100% | +0% | 2026-06-30 |
| game-visuals | Visual design principles for legibility, color hierarchy, and art direction in games | 33% | 100% | +67% | 2026-06-30 |
| level-design | Combat arena and action game level design: spatial choice, legibility, encounter flow | 20% | 100% | +80% | 2026-07-23 |
| llm-inference-serving | Production LLM inference: parallelism strategy, speculative decoding, KV cache management, serving architecture, and cost optimization | — | — | — | — |
| love2d-fennel | Fennel + Love2D interactive development: REPL workflow, hot-reloading, mode architecture | 20% | 87% | +67% | 2026-07-23 |
| ml-platform-architecture | ML platform architecture: maturity models, cluster topology, GPU admission control, migration frameworks, and platform engineering patterns for production ML | — | — | — | — |
| ml-post-training | SFT and RL post-training mechanics: data pipelines, reward design, LoRA, debugging | 64% | 84% | +20% | 2026-07-23 |
| multi-agent-collaboration | Designing multi-agent topologies, context sharing, handoff protocols, and failure modes | 47% | 93% | +47% | 2026-07-23 |
| nix | NixOS, Home Manager, nix-darwin, flakes, devShells, and declarative system configuration | 62% | 70% | +9% | 2026-07-23 |
| nix-dendritic | Aspect-oriented flake-parts Nix configuration for multi-host, multi-platform setups | 49% | 90% | +41% | 2026-07-23 |
| odin-design | Idiomatic Odin patterns, allocators, package structure, and LLM knowledge-gap corrections | 67% | 86% | +19% | 2026-07-23 |
| odin-gamedev | Odin game architecture with Raylib/Sokol: entity management, hot reloading, game state | 71% | 89% | +18% | 2026-07-23 |
| physical-ai-systems | Data pipelines, inference patterns, and training optimization for AI systems operating in the physical world | — | — | — | — |
| python-design | Python-specific design patterns, idioms, type choices, and anti-patterns | 71% | 98% | +27% | 2026-09-09 |
| rag-design | Retrieval pipeline design: chunking, embeddings, hybrid retrieval, and structured indexes | 40% | 93% | +53% | 2026-07-23 |
| rl-generalization | Designing RL training curricula that produce broad, transferable capability rather than narrow skills | 60% | 96% | +36% | 2026-08-18 |
| software-design | Core software design principles: composition, minimal interfaces, and clean boundaries | 77% | 85% | +8% | 2026-09-09 |
| test-design | Test quality tradeoffs using Kent Beck's Test Desiderata: behavior over structure | 67% | 100% | +33% | 2026-06-30 |
Skills that provide domain facts and syntax the model already knows; they exist for consistent routing and conventions.
| Skill | Summary | Baseline | With Skill | Δ | Last Run |
|---|---|---|---|---|---|
| docker | Dockerfile authoring, multi-stage builds, image optimization, and container networking | 78% | 100% | +22% | 2026-07-23 |
| game-rendering | 3D rendering pipeline math: coordinate spaces, rasterization, projection, lighting | 67% | 100% | +33% | 2026-06-30 |
| godot | Godot 4.x GDScript development: scenes, nodes, signals, AutoLoads, and patterns | 60% | 93% | +33% | 2026-07-22 |
| godot-shader | Godot 4.x GDSL shader authoring: spatial, canvas_item, and particle shaders | 40% | 73% | +33% | 2026-07-23 |
| gpu-rendering-architecture | Modern GPU programming model: bindless resources, GPU-driven rendering, synchronization | 53% | 67% | +13% | 2026-07-23 |
| grafana | Grafana dashboard authoring, template variables, IaC provisioning, and alerting | 47% | 87% | +40% | 2026-07-23 |
| helm | Helm chart installation, value overrides, upgrades, rollbacks, and chart discovery | 70% | 83% | +13% | 2026-07-23 |
| homelab-monitoring | Deploying Loki, Prometheus, Grafana, and Alertmanager on k3s or Docker Compose | 47% | 87% | +40% | 2026-07-23 |
| k3s | Installing and managing self-hosted Kubernetes clusters with k3s | 61% | 94% | +33% | 2026-07-23 |
| k8s-networking | DNS, Ingress, TLS, NetworkPolicy, and VPN access for self-hosted Kubernetes | 60% | 88% | +28% | 2026-07-23 |
| k8s-operations | Day-2 Kubernetes operations: probes, rollouts, hardening, and deployment automation | 40% | 60% | +20% | 2026-07-23 |
| k8s-storage | Kubernetes ConfigMaps, Secrets, PersistentVolumes, and backup strategies | 60% | 87% | +27% | 2026-07-23 |
| k8s-workloads | Kubernetes workload resources: Deployments, StatefulSets, DaemonSets, Jobs, Services | 67% | 80% | +13% | 2026-07-23 |
| logql | LogQL query syntax for Grafana Loki: filtering, parsing, and log-based metrics | 33% | 93% | +60% | 2026-07-23 |
| love2d | Love2D 11.x Lua development: callbacks, input, shaders, and game loop patterns | 60% | 87% | +27% | 2026-07-23 |
| nix-packaging | Writing Nix derivations with stdenv and language-specific builders from scratch | 60% | 100% | +40% | 2026-07-23 |
| promql | PromQL query syntax for Prometheus: rates, histograms, alerting, and capacity planning | 78% | 100% | +22% | 2026-07-23 |
Skills not yet assigned to a section.
| Skill | Summary | Baseline | With Skill | Δ | Last Run |
|---|---|---|---|---|---|
| ray-data-pipelines | Design patterns for building production Ray Data pipelines across CPU/GPU heterogeneous clusters | — | — | — | — |
| ray-kubernetes-infrastructure | Production patterns for running Ray on Kubernetes with topology-aware scheduling, fault tolerance, and full-stack observability | — | — | — | — |
<skill>/SKILL.md — skill definition + references
evals/<skill>/ — eval definitions + benchmarks
scripts/ — tooling (generate_readme.py)