Agent skills for Claude Code, Gemini CLI, and any agent that follows the agentskills.io spec.
- tidy-code — Review within-file code smells across any programming language. Flags hidden dependencies, god functions, deep function nesting, silent failures, poor naming, and eight other named principles.
- tidy-project — Read-only, three-phase audit of repo-level structure for easier solo-dev maintenance. Flags dead code, duplication, premature abstraction, over-nested folders, misnamed files, config sprawl, scattered cohesion, vestigial framework scaffolding, over-coupled files, and docs/code drift.
- plain-language — Review text for plain-language compliance based on the U.S. federal government Plain Language Guidelines.
Every skill produces a single self-contained LLM-ready plan at docs/YYYY-MM-DD-<skill-name>-plan.md (or the repo root if no docs/ directory exists). The orchestrator surfaces only that path to the user. Intermediate scratch artifacts (per-phase outputs, adversarial reviews) live under .agents/local/state/<skill-name>-<runid>/ and exist only to support Resume on a crashed run — they are not surfaced unless you explicitly ask where they live. <runid> is a UUIDv7 (generated by each skill's scripts/uuidv7.sh), so concurrent runs — the same skill in several checkouts at once, or a fan-out of runs — never share a state directory or clobber each other. See repo-root STANDARDS.md § Run-ID scoping.
Progress for each skill is tracked via your CLI's built-in task list — TaskCreate on Claude Code, write_todos / tracker_* on Gemini CLI, update_plan on Codex, or a re-printed markdown checklist on any other host. You can watch phases tick from pending → in_progress → completed in realtime. Each skill ships references/progress-tracking.md with the per-host adapter rules.
# Install a skill
npx skills add ggwicz/skills --skill <skill-name>
# List available skills in this repo
npx skills add ggwicz/skills --list