Agent Scripts

A public collection of reusable agent workflows and prompts.

To install a skill, point your agent to the skill you want and ask it to install it.

Skills

• /agent-browser — automate browser, website, and Electron app workflows with the agent-browser CLI.
• /autoreview — run structured Codex or Claude code reviews before commit or ship.
• /brainstorm — generate and explore ideas before choosing a direction.
• /codebase-quiz — turn real implementation work and prior coding sessions into a visual HTML lesson with an interactive quiz.
• /create-cli — design command-line interfaces, flags, help, output, errors, config, and dry-run behavior.
• /frontend-design — build distinctive, polished frontend UI instead of generic layouts.
• /grill-me — interview the user relentlessly to stress-test a plan, design, or approach.

Each one lives in skills//SKILL.md.

Skills are reusable workflows. Matt Pocock’s convention is: user-invoked skills set disable-model-invocation: true; model-invoked skills omit that field and keep a model-facing description.

Prompts

• prompts/security-audit.md — phased, report-only security audit prompt.

Prompts are manual, explicit-use instructions. They live outside skills/ so agents do not invoke every workflow by themselves.

Codex CLI does not currently use disable-model-invocation to control skill invocation. For Codex, implicit skill use is controlled by agents/openai.yaml with policy.allow_implicit_invocation; keep Codex-only manual workflows in prompts/.

Acknowledgements

This collection builds on ideas from Matt Pocock and Brian Madison’s agent workflow methodology. I love their work.

• /agent-browser is from Vercel Labs’ agent-browser.
• /autoreview and /create-cli are from Peter Steinberger’s agent-scripts.
• /frontend-design is from Anthropic’s skills.
• /grill-me is a standalone adaptation of Matt Pocock’s grill-me and grilling skills.
• /brainstorm is an original standalone workflow inspired by BMAD’s brainstorming approach.