Skills

How I work with AI agents: what they read, the steps they follow, how I set up a project, and the skills I use. Each skill links to its source.

How I work

Source and workflow

Two pieces. The source is what stays still: what the agent reads and what it can use. The workflow is what moves: a line of steps, each one taking from the source what it needs.

Source

StaticText the agent reads. Lives in Git.

  • AGENTS.md
  • Docs
  • Specs
  • Rules
  • Skills
  • Agents
  • Memory

ToolsWhat the agent can use. Connected per session, with logins.

  • MCP · Plane, Jira
  • MCP · Figma
  • CLI · gh, vercel
  • Browser
  • Hooks · CI
  1. Routing
  2. Research
  3. Spec & plan
  4. Implement
  5. Validate
  6. Deliver
  7. Measure
Workflow
Each step takes from the source. What it learns improves it.

How I build with AI

  1. Research

    Web search, codebase, project docs, second brain. Check costs. Reproduce the bug.

  2. Spec & plan

    plan mode

    Grill questions, A/B/C variants, prototype, scope. Write spec.md, plan.md, Gherkin.

  3. Implement

    UI states, tests, lint, typecheck, build. Agent code review.

  4. Validate

    agent-browser or Playwright. Gherkin with evidence. 1440, 375, dark.

  5. Deliver

    Small commits, PR with evidence, Vercel preview, QA handoff.

PrinciplesPer-project harness (AGENTS.md, rules, skills, MCPs). Small scope. Traceability. Parallel subagents.

How I set up a project

Every chat starts knowing nothing, whether it is Claude, Codex or Cursor. So the repo explains itself in plain files any agent reads. A session is three things: the chat, its connections, and the files.

  • project/
  • ├── AGENTS.mdmap, commands, limits (~100 lines)
  • ├── docs/the truth
  • │ ├── product/what and for whom, glossary
  • │ ├── architecture/how it is built today
  • │ ├── decisions/ADRs: why
  • │ ├── design/DESIGN.md, tokens
  • │ └── work/<ID>-slug/one ticket
  • │ ├── spec.mdwhat, why, Gherkin
  • │ ├── plan.mdhow, in small steps
  • │ ├── handoff.mdwhere it stopped
  • │ └── results.mdevidence
  • ├── .agents/neutral, any agent
  • │ ├── skills/procedures
  • │ ├── rules/by path
  • │ └── connections.mdwhich MCPs, no secrets
  • ├── scripts/own tools as CLIs
  • ├── .github/workflows/CI nobody skips
  • └── .claude/ .cursor/adapters, generated

Skills

Research & data

scrape-catalog@franciscobeccariaSweeps and filters product listings.x-reader@franciscobeccariaReads your X bookmarks with your Chrome session.

Writing

humanizer@bladerRewrites AI-sounding prose.

Frontend

emil-design-eng@emilkowalskiUI polish and motion taste.animate@emilkowalskiBuilds animations from scratch.review-animations@emilkowalskiReviews motion against rules.improve-animations@emilkowalskiAudits and plans animation fixes.find-animation-opportunities@emilkowalskiFinds where motion helps.animation-performance@emilkowalskiFixes animations that drop frames.animation-vocabulary@emilkowalskiWords for the motion you want.apple-design@emilkowalskiApple UI and motion principles.break-ui@emilkowalskiStress-tests UI with worst-case data.prototype@emilkowalskiBuilds UI variants to compare.pick-ui-library@emilkowalskiPicks a library for the task.mobile-native@emilkowalskiMakes web apps feel native on phones.

Workflow

pstack@cursorRigorous multi-agent engineering in Cursor.grill-me@mattpocockGrills a plan until questions are settled.research@mattpocockResearches primary sources in the background.to-spec@mattpocockTurns the chat into a published spec.implement@mattpocockImplements from a spec or tickets.code-review@mattpocockReviews a diff for standards and spec.