Skills

A personal collection of Agent Skills — reusable instructions that teach coding agents (Claude Code, Codex, and other Agent Skills–compatible harnesses) how to do specific tasks.

Inspired by Matt Pocock's skills and his "AI Builder" way of working: plug-and-play skills for real engineering, not vibe coding.

Each skill lives in its own folder under skills/ with a SKILL.md that the agent loads on demand.

Install as a Claude Code plugin

This repo is packaged as a Claude Code plugin (see .claude-plugin/). Add the marketplace and install:

/plugin marketplace add Yoliani/skills
/plugin install yoliani-skills@yoliani

Install via symlinks

Symlink every skill into the local skill directories used by each agent harness:

./scripts/link-skills.sh

This links each skill into:

• ~/.claude/skills — Claude Code
• ~/.agents/skills — Codex and other Agent Skills–compatible harnesses

Because each entry is a symlink into this repo, a git pull is all that's needed to keep installed skills up to date.

To list every skill in the repo:

./scripts/list-skills.sh

Reference

Engineering

Skills for daily code work.

• crabbox-setup — Install and configure crabbox, the remote execution control plane that leases cloud machines, syncs your repo, and runs commands remotely. Covers CLI install, broker login, doctor checks, and scaffolding a repo-local .crabbox.yaml.
• crabbox-usage — Day-to-day crabbox workflows: one-shot remote runs, warm boxes for repeated runs, SSH/attach, file copy, artifacts, port forwarding, screenshots/VNC, and cost hygiene.
• agents-md — Write, audit, or rewrite AGENTS.md files so they carry only what an agent can't infer from the repo.
• loop-design — Design the task DAG a work loop will walk: turn a spec, plan, or issue list into a .agent-loop/<effort>/ directory with tasks, dependencies, topology decision, and durable shared state files.
• loop-operate — Run the work loop over an existing .agent-loop/ task DAG: claim ready tasks, execute, verify, update shared state, and react to runtime reorg triggers.
• why — Code archaeology: recover the design rationale behind code by fanning investigators out across every available evidence source (source control, tickets, docs, chat, observability, error tracking, analytics) and returning a cited, confidence-calibrated read.

Productivity

General workflow tools, not code-specific.

• hyperplan — Interrogate a task's requirements until every assumption is explicit, then generate an implementation-ready plan for a build agent.
• show-me-your-work — Keep a reviewable decision trail for long-running or unattended work: a TSV log with one row per decision (what, why, evidence, result).

Personal

Tied to my own setup and voice.

• unslop — Remove AI tells from prose and restore human voice, preserving meaning and intended tone.
• pixel-art-creator — Create new pixel art sprites from scratch: canvas, layers, and basic drawing primitives.
• pixel-art-animator — Build sprite animations with frames, animation tags, and onion skinning.
• pixel-art-professional — Apply advanced techniques: dithering, palette optimization, anti-aliasing, outlining.
• pixel-art-exporter — Export sprites to PNG, GIF, or spritesheets with JSON metadata for game engines.

The four pixel-art-* skills are copied from willibrandon/pixel-plugin (MIT) and require its Aseprite MCP server to be configured.