roger-skills

A curated collection of reusable skills for AI coding agents, distilled from real-world development and engineering workflows.

What is a skill?

A skill is a reusable prompt playbook stored as SKILL.md that an agent executes on demand. Skills follow the agentskills.io open standard and are compatible with Claude Code and OpenAI Codex CLI.

Invoke with a slash command: /commit, /debug my-file.go, /pr-review 42

Skills

Skill	Trigger	Description
commit	/commit	Stage changes and write a Conventional Commit message
create-pr	/create-pr	Create a PR with title, description, and test plan
pr-review	/pr-review [PR]	Review a PR for correctness, security, tests, and style
security-review	/security-review	Audit code for OWASP Top 10, secret leaks, and insecure patterns
debug	/debug	Systematically diagnose a bug and fix the root cause
refactor	/refactor [file]	Improve code clarity without changing behavior
simplify	/simplify [file]	Remove unnecessary complexity — dead code, over-abstraction, YAGNI
test-gen	/test-gen [file]	Generate unit tests covering happy path, edges, and errors
docstring	/docstring [file]	Add or improve docstrings and inline comments
changelog	/changelog [tag]	Generate a Keep a Changelog entry from recent commits
migrate	/migrate [path]	Audit database migrations for lock risks, data loss, and zero-downtime safety
dep-update	/dep-update	Analyze outdated dependencies and produce a prioritized batch upgrade plan
postmortem	/postmortem	Generate a blameless incident postmortem with timeline, RCA, and action items
adr	/adr	Create an Architecture Decision Record
spec	/spec	Write a requirements specification with acceptance criteria

Installation

Copy the skills you want into your project or home directory:

# Project-level (affects this repo only)
cp -r skills/commit .claude/skills/

# Personal (affects all projects)
cp -r skills/commit ~/.claude/skills/

Design guidelines

Follow these principles when adding or editing a skill:

• Process-driven, not prompt-driven. Each skill defines an explicit step-by-step workflow, not a vague "help me do X". Steps are ordered and checkable.
• Constraints first. Hard limits (what the skill must never do) belong in the skill itself, not in a separate global config. This makes skills portable.
• Minimal tool permissions. allowed-tools grants only the specific commands the skill needs — not broad Bash(*).
• Triggerable by natural language. The description field covers realistic phrasings a developer would actually use, so the agent can auto-invoke without a slash command.
• Single responsibility. One skill, one workflow. If a skill is doing two separate things, split it.
• Language-agnostic where possible. Skills that work across multiple languages (debug, refactor, docstring) should detect the project's language rather than hardcode one.