Never defer your thinking to AI

Skills for Claude Code that keep my judgment in the loop and my voice in the writing.

Install

Any agent

Installs the skills you pick into Claude Code, Codex, Cursor, Amp, or any other agent the skills CLI supports. Nothing runs in the background unless you register the hooks: on Codex, one command per plugin adds them to ~/.codex/hooks.json. You invoke a skill when you want it, and the run commands below drop their plugin prefix on this route.

npx skills add craigmdennis/design-skills

Claude Code

Add the marketplace once, then install any skill below with its own command. The plugin route also registers the hooks a skill ships with: Field Notes captures on its own in a project you've asked it to track, and Writing re-applies its conversation checks on every turn.

/plugin marketplace add craigmdennis/design-skills
  • Field Notes

    Capture decisions and context as you work, so the write-up is already half done.

    Git history records what was built. field-notes records why. The agent maintains a decision log every session (the problem, the decisions made, the approaches rejected) as the source for a post, case study, or retro. Nothing is committed. The log is an untracked .field-notes/ folder, and a session-start hook gives the agent its instructions. Three more hooks log prompts verbatim, collect insights, and ask once a session for the reason for a decision. Every hook does nothing until a project is tracked. On Codex, one command registers three of the four hooks; insight capture stays Claude Code only. I use it on every project.

    /plugin install field-notes@design-skills

    Run it/field-notes:track-project

    View Field Notes on GitHub

  • Writing

    Three writing standards, one per genre, so the agent stops writing like an AI in replies, documentation, and anything published under an author's own name.

    Each genre has its own skill, because the rules that improve one make the other two worse. conversation-prose governs the agent's replies. A hook re-applies its seventeen checks on every turn, so the standard is still applied at turn 90. documentation-prose governs skill files, READMEs, specs, and pull request bodies. published-prose governs posts, case studies, and resume copy, from a voice profile built by interview on first use. All three fail mannered prose, metaphor or flourish where a direct statement exists, a rule taken from Anthropic's writing-density guidance. On Codex, one command registers the same per-turn hook. A blinded harness measures the result: violations per 1,000 words decrease by 52% in replies and by 84% in documentation.

    /plugin install writing@design-skills

    Run it/writing:published-prose

    View Writing on GitHubRead how I built and measured them

  • Design Goal Setting

    Interview-driven goal-setting and performance-review prep for designers.

    Interviews you one question at a time, maps your evidence to a behavioural bar, and picks the moves that close the gaps. Fills a bundled review template; nothing leaves your machine.

    /plugin install design-goals@design-skills

    Run it/design-goals:set-goals

    View Design Goal Setting on GitHub

  • Design Impact Report

    Turn one initiative you owned into a short, shareable impact report.

    Interviews you about a single piece of work you owned, then fills a bundled impact-report template you can hand to your manager or bring to a review.

    /plugin install impact-report@design-skills

    Run it/impact-report:write-report

    View Design Impact Report on GitHub

  • Strengthen Case Study

    Strengthen an existing portfolio case study through an evidence-first interview.

    For the case study that reads like a PM writeup, opens at the solution, or makes claims it never proves. Interviews you one question at a time, then trims and corrects your own words rather than writing over them. Every gap becomes a question you answer, never a placeholder left in your draft. Works on a file, a URL, or pasted text.

    /plugin install case-study@design-skills

    Run it/case-study:strengthen

    View Strengthen Case Study on GitHub