Dev & Coding

A design-first workflow skill that sketches types, signatures, and module boundaries before implementation, then treats that sketch as a contract while the code fills in.

A design workflow skill that sketches types, signatures and module boundaries first, synthesizes multiple model perspectives, then fills in code against the agreed sketch.

Hands-on skill for installing, running and benchmarking Mamba-1/Mamba-2 linear-complexity SSMs.

A skill that walks you through loading, fine-tuning (LoRA/QLoRA), pretraining and deploying 20+ LLM architectures with Lightning AI's LitGPT.

A skill that walks Claude through configuring and launching large-scale LLM pretraining with torchtitan — 4D parallelism (FSDP2/TP/PP/CP), Float8, and distributed checkpointing from 8 to 512+ GPUs.

Delegates a task to another AI coding CLI, or fans one prompt out to several models in parallel and synthesizes their independent opinions.

A workflow skill that decomposes repetitive, large-scale code changes into independent units and runs one parallel agent per isolated git worktree.

Gives Claude battle-tested Compose Multiplatform rules for expect/actual, shared resources, iOS/Android interop and cross-target performance.

A deep reference for advanced Jetpack Compose visuals: M3 Expressive spring motion, AGSL shaders, and Canvas generative drawing with correct SDK/version gating.

A QA harness that exercises the omo Codex plugin in a real codex app-server using an isolated CODEX_HOME and a local mock model, with hook firing proven by notifications.

A rulebook for adding or extending Paperclip's full-stack Product E2E eval workflows, fixtures, matchers, and evidence paths for local or Daytona runs.

An installer skill that wires an Emacs channel into NanoClaw over a local HTTP bridge, giving you a chat buffer and org-mode integration inside Emacs.

Installs OpenAI Codex (app-server) as a first-class agent backend for NanoClaw, switchable per group.

Wires a stdio MCP server into the NanoClaw agent-runner so the containerized agent can call local models served by the Atomic Chat desktop app.

Scans a TypeScript project for unused code, confirms each candidate via LSP, then removes it through parallel per-file batches with atomic commits.

Walks a directory tree and auto-generates or repairs EGM-compliant READMEs documenting each package's interface, invariants, and side effects.

Installs, repairs and tests Claude Code hooks so risky commands are blocked mechanically instead of by polite instructions.

A BMad-native skill that slices initiatives into epics and stories, writes/refines tickets, and runs the whole board.

A pre-npm-publish gate that splits unpublished changes into up to 10 groups for parallel deep review, adds a 5-agent holistic pass, and an oracle release synthesis.

Turns a vague or large request into ONE decision-complete work plan: explores the codebase first, asks only the forks it cannot resolve, and writes the plan only after your explicit approval.