skills

A collection of personal Claude Code skills - reusable prompt shortcuts for common workflows.

Prerequisites

• Claude Code CLI and/or OpenAI Codex CLI installed

Install

git clone https://github.com/pstaylor-patrick/skills.git
cd skills
./install.sh

This creates symlinks for all skills, making them available system-wide:

• Claude Code: ~/.claude/commands/{name}.md (file symlinks)
• OpenAI Codex: $CODEX_HOME/skills/{name}/ (directory symlinks, defaults to ~/.codex/skills/)

Re-run ./install.sh any time you install a new CLI to pick it up. Restart Codex after installing new skills so it reloads them.

Codex skills are not invoked as slash commands. In Codex, mention the skill name in your prompt, for example: Use f3:push to push this branch and validate the PR.

Uninstall

./install.sh --uninstall

Removes symlinks from both ~/.claude/commands/ and $CODEX_HOME/skills/.

Skills

/f3:auto

High-autonomy orchestrator that turns rough, unstructured prompts into review-ready pull requests. Asks up to 3 clarifying questions, then runs autonomously through implementation, slop cleanup, quality gates, preflight code review, push/PR creation, QA, and a final autofix review pass. Opens the finished PR in your browser.

/f3:auto "add an org invitation flow"
/f3:auto "fix the broken dashboard filtering and get this branch ready for review"
/f3:auto "take this half-finished work and make the PR fully review-ready"

/f3:next

Assess the current state of your work and get one opinionated recommendation for the best next step. Reads git state, GitHub PR status, and project context to tell you THE answer, not a menu of options.

/f3:next
/f3:next --verbose
/f3:next --why

/f3:code-review

Code review with worktree-isolated fix verification. Every finding is validated by applying the suggested fix in an isolated worktree and running quality gates - findings that break the build are dropped. Supports GitHub PR reviews, local-only output, autonomous auto-fix, and multi-round sweep mode.

/f3:code-review 42
/f3:code-review https://github.com/owner/repo/pull/42
/f3:code-review --local
/f3:code-review --autofix
/f3:code-review --sweep

/f3:demo

Generate a reusable demo/QA runbook from the current feature branch. Analyzes code changes, commits, and PR context to create a step-by-step walkthrough saved as a skill in the target repo's .agents/skills/ directory. Usable for both QA testing and stakeholder Loom demos.

/f3:demo
/f3:demo --update
/f3:demo --dry-run

/f3:push

Auto-commit, push the current branch, ensure a PR exists against the default branch, and refresh the PR title and description to reflect all changes on the branch. Then autonomously validate every unchecked test-plan checkbox via terminal commands (build, lint, typecheck, test) and code-level checks -- no browser automation. Posts a validation comment and checks off passing items.

/f3:push
/f3:push --dry-run

/f3:rebase

Rebase the current branch onto a base branch with automatic conflict resolution, Drizzle migration cleanup, and force-push. Infers the base branch from the current PR or falls back to the repo default. Automatically removes all Drizzle database migrations from the feature branch (you regenerate them manually via Drizzle Kit after). Asks for user input only when a conflict is genuinely ambiguous.

/f3:rebase
/f3:rebase main
/f3:rebase develop --no-push
/f3:rebase --dry-run

/f3:qa

Autonomous QA testing that synthesizes test plans from PR context and code diffs, then executes via browser automation (Playwright MCP or CDP). Auto-judges pass/fail and posts evidence to the PR. Use --guided for interactive human-driven testing.

/f3:qa 42
/f3:qa https://github.com/owner/repo/pull/42
/f3:qa --guided
/f3:qa --post-merge

/f3:resolve-threads

Address every unresolved conversation on a GitHub PR. Fetches all review threads, top-level comments, and review summaries, then classifies each (filtering out bots, CI previews, and redundant review summaries). For actionable feedback: tests a fix in an isolated worktree, and if it passes quality gates, squash-merges it into the branch and replies confirming. For inapplicable suggestions: replies with reasoning. Resolves all threads when done. Top-level comment replies include a blockquote of the original for clarity.

/f3:resolve-threads
/f3:resolve-threads 42
/f3:resolve-threads https://github.com/owner/repo/pull/42
/f3:resolve-threads --dry-run

/f3:react-refactor

Extract business logic from React/Next.js components into tested custom hooks. Uses Vercel react-best-practices (64+ rules) as the industry baseline, layered with opinionated architecture preferences: hooks in *.ts files, comprehensive vitest coverage, zero eslint-disable, named exports only.

/f3:react-refactor src/components/Dashboard.tsx
/f3:react-refactor --all
/f3:react-refactor --branch feature/new-dashboard
/f3:react-refactor --dry-run

Vercel react-best-practices (64+ industry rules) is installed automatically by ./install.sh via the skills CLI. The skill degrades gracefully if the external dependency is missing.

/f3:slop

Scan for and remove common AI-generated slop from your branch changes (default) or the entire repo (--repo). Detects em dashes, excessive documentation, disabled quality gates, band-aid exclusions, over-complicated abstractions, dead code, error theater, and type safety escapes. Interactive by default - presents findings and asks before fixing.

/f3:slop
/f3:slop --repo
/f3:slop --dry-run
/f3:slop --auto

/decide-for-me

Tells Claude to pick the best approach instead of presenting options. Evaluates simplicity, reliability, scalability, maintainability, and end-user experience.

/decide-for-me

/spec-gen

Launches an in-depth interview to build a complete implementation spec. Covers technical details, UI/UX, concerns, tradeoffs, and scope - asks non-obvious questions until the spec is ready.

/spec-gen
/spec-gen "Add org invitation flow"

/validate-quality-gates

Runs build, lint, typecheck, test, and test:coverage in a loop - fixing failures as it goes until all checks pass cleanly.

/validate-quality-gates

License

MIT