Practical agent skills for real engineering work with Claude Code, Codex, Gemini CLI, and other tools that support the Agent Skills format.
These are the skills I keep reusing across repos because they solve recurring failure modes: weak plans, lost context, shallow reviews, and gradual repo drift.
npx skills@latest add andrewcodesit/skillsPick the skills you want from the interactive prompt. Works with Claude Code, Codex, Cursor, Windsurf, and other agents that support the format.
Install a specific skill without the picker:
npx skills@latest add andrewcodesit/skills --skill code-reviewAn agent will happily write a plan and then quietly deviate from it the moment implementation gets hard. plan forces a real plan to exist before code gets touched. execute runs that plan and then checks the result against it - so drift gets caught, not shipped. plan-review catches bad plans before they become bad code. Sometimes I even run plan-review multiple times in a row, iterating on the plan until it's solid before I let it near the codebase.
Every new session starts from zero unless you've documented the repo for the agent. init-context-files gives agents a persistent context/ directory instead of re-deriving the codebase from scratch each time. handoff captures exactly where a session left off so the next one - yours or another agent's - can pick it up cold.
Agents reviewing agent-written code tend to either approve everything or flood you with nitpicks. code-review is tuned to focus on real bugs and real risks, not style noise.
The other half of the problem is when review happens. An agent that reviews its own work seconds after writing it, with the plan still in context, mostly re-confirms its own assumptions - so execute verifies the implementation against the plan and stops there, and the real review is a separate pass over the finished diff. Whether that's code-review or an automated reviewer on the merge request is up to your repo; either way it's offered, never forced.
Dead code, leftover debug logs, and unused imports pile up quietly across sessions. cleanup scans for it, cites file:line, and fixes it category by category with your confirmation - not a silent mass rewrite.
Starting from a vague idea is where agents lose the thread fastest. define-spec turns the idea into a concise spec, breakdown-tasks turns that spec into local or external tasks with epics when the work is large, git-publish puts the branch up as a draft merge/pull request once there's something worth showing, and close-task handles the finish line: final validation, marking that request ready for review, routing whatever the review turns up, and task completion.
- init-context-files - Creates a
context/directory with structured markdown that gives agents persistent project knowledge.
- cleanup - Scans the repo for dead code, debug artifacts, and other cruft, then fixes it category by category with confirmation.
- code-review - Deep code review focused on structural issues, behavioral risks, and missing coverage.
- execute - Runs an approved plan, then verifies the implementation against the plan, repo standards, and the contracts the plan named.
- git-publish - Commits the current branch, pushes it, and opens a draft merge/pull request on GitHub, GitLab, or Azure DevOps.
- verify-ui - Verifies local UI changes in a running app, including the browser flow, layout, and visible regressions.
- verify-ts - TypeScript strictness audit enforcing no-any, unknown at boundaries, discriminated unions, branded types, exhaustiveness checks, and runtime validation.
- breakdown-tasks - Breaks specs or project goals into local or external tasks, with epics for larger work.
- close-task - Finishes a task by running final validation, marking the merge/pull request ready for review, routing the resulting review findings, and marking the task done.
- define-spec - Turns a project idea, feature request, bug, or vague goal into a concise implementation-ready spec.
- plan - Writes an implementation plan for a feature, task, or spec.
- plan-review - Reviews an implementation plan and critiques it before execution.
- handoff - Generates a comprehensive handoff spec so another agent can pick up the session.
Each skill is a SKILL.md file with YAML frontmatter (name, description) plus instructions the agent follows when it matches your request, grouped under skills/<category>/<skill-name>/SKILL.md.
Skills follow the Agent Skills specification and are compatible with any agent that supports it.
This repo is intentionally simple:
- Skills live at
skills/<category>/<skill-name>/SKILL.md. - Every skill must include YAML frontmatter with a non-empty
nameanddescription. - Validation is handled by
scripts/validate-skills.js. - Test the repo with
npm test. - There is no build step, transpilation, or bundler.