Playbooks I wrote for my own coding agents and kept using. Each one is a
plain SKILL.md file, so it works in Claude Code, Codex, and most other
agents. Every entry below installs with one line.
Claude Code
/pluginmarketplaceaddadityaarakeri/skills
Adds the whole catalog to /plugin so you can browse and pick. Other agents: use the per-skill npx skills lines below.
7 plugins · 16 skills
Engineering discipline
Make a coding agent work like a careful senior engineer: read the repo first, test before claiming done, and leave a trail the next session can follow.
senior-agent-skills
10 skills
Makes a coding agent plan, test, and verify before shipping.
Ten portable playbooks covering the whole loop: understand the repo, plan risky work, build test-first, debug scientifically, refactor safely, audit dead code, review the diff, prove completion, and keep git history clean.
Goal-driven autonomous execution. The agent inspects the repo, picks the lowest-risk path, makes routine scope and naming calls itself, verifies the result, and reports once at the end. Platform safety gates still apply.
Audits a repo, then fixes it one test-first item at a time, across context resets.
Builds a checklist of quality, security, functionality, CI, and performance issues and works through it with strict TDD. State lives in CHECKLIST.md and .agent/, so a fresh session picks up exactly where the last one stopped.
Output that reads and looks like a person made it.
humanize-content
1 skill
Strips AI tells from text without adding new ones.
Rewrites robotic text in one specific person's voice. It never invents numbers, anecdotes, or opinions; gaps come back as [FILL: ...] prompts. Ships a zero-dependency scanner you can run in CI.
The bundled scanner run against the repo's own test fixtures
For docs, API references, runbooks, and portfolios. Picks from contrast-tested themes, ships light and dark mode and mobile layouts by default, and audits its own output against a list of AI design tells before you see it.
For React, Next.js, Vue, Svelte, and Astro work in a real codebase. Reads the project first, sets design tokens before components, then checks accessibility and Core Web Vitals before calling it done. For static docs pages, use handcraft-html.
For the decisions that do not come with a test suite.
judgment-coach
1 skill
Coaches a decision instead of making it for you.
Asks for your view first, separates evidence from assumption, finds the one uncertainty that could flip the call, and designs the cheapest test for it. Ends with a decision memo and a review date.