Essential prompt library
Skill packs and prompts are provided in English so they work best with the AI tools. Use them as-is.
This is your working library: fifteen prompts ready to copy, fill in what's between [brackets], and run. They go from simple to powerful — first seven general prompts that work from day one, no skills or connections needed; then eight engineering recipes that run the full professional workflow.
The prompts are written to paste as-is into Claude Code, which works best in English. The title and the line above each one tell you what it does.
Want them all in one file? Download the full library at prompts-esenciales.md.
General prompts
They work without skills or connections: copy them, change what's in brackets, and use them.
1. Give it permanent context about you
Interview me for 10 minutes to understand my business, my voice and my goals. With my answers, create context files in the context/ folder that you can read every session. Never make up data: only use what I tell you.
2. Create a new skill
Create a skill for [describe the task you repeat often]. So that when I type
/[short-name] it always follows the same format. Ask me whatever questions you
need to build it well.
3. Turn a document into something actionable
Here is [paste or attach the document]. Extract the key points, tell me what decisions it demands of me, and build a concrete, prioritized task list.
4. The anti-fabrication rule (paste it when precision matters)
Rule that overrides everything else: never invent a figure, an event, or a quote that doesn't come from the sources I gave you. If you don't have it, tell me — don't fill it in.
5. Content repurposing
Take this [video / article / transcript] and create for me: a LinkedIn post, a short email for my list, and 3 social post ideas. All in my voice (check my context/ files first).
6. Research with judgment
Research [topic] from the last 30 days. Give me: what's working, what everyone is saying that I could contradict with data, and where there's high demand with low supply. Cite your sources.
7. Review with sub-agents (advanced)
Generate [whatever], then spin up a sub-agent that reviews the result against my voice rules in context/. Have it tell me what it corrected and why.
Engineering prompts
These eight recipes use the skills you installed in the skills lessons and run, step by step, the professional workflow from The Big Idea — the 7 rules. Adapted, with credit, from Matt Pocock's engineering workshop.
The ritual before each one: 1) new session + repo · 2) set the branch · 3) pick the model · 4) click Copy, paste, fill the [brackets], press Enter. (Branches and models are covered in the lesson Your first session, branches & models.)
grill-me — Opus 4.8 · branch plan/<feature> · in Plan mode
Get aligned before building.
Use the grill-me skill. I want to build: [describe your idea in one or two sentences]. Interview me one question at a time — each with your recommended answer — to expose unclear requirements, hidden assumptions, risks and edge cases. Don't move on until we're aligned. When we finish, summarize the shared design concept we agreed on.
Real example: what a grill-me session actually looks like
Your client Sarah Chen sends a vague Slack: "Students drop off after a few lessons — let's add some gamification." Watch grill-me turn that fog into a clear plan, one question at a time.
You: Use the grill-me skill. Sarah wants gamification on our course platform to improve retention. Grill me.
Claude: Q1 · Points economy. What actions earn points, and how much? Recommendation: only lesson completions and quiz passes to start. Skip video views (easy to game).
You: Agree. Lessons and quizzes only.
Claude: Q2 · Retroactive? Should past completions earn points too? Recommendation: no backfill — start fresh at launch.
You: Good call. No backfill.
Claude: Q3 · Where does it show? Dashboard, lesson page, or both? Recommendation: the dashboard first — visible right after login.
You: Dashboard. Ship it.
Claude: Shared design concept. Points for lesson + quiz completions only · no retroactive backfill · simple level curve · shown on the dashboard. Ready to turn this into a PRD?
That agreed picture — not a fancy document — is the real asset. Next: run to-prd, then to-issues.
to-prd (write a PRD) — Opus / Sonnet · branch plan/<feature>
Summarize the alignment into a destination document.
Use the to-prd skill. Based on our grilling session, write a one-page PRD: problem, user stories each with a definition of done, a proposed module map, testing decisions, and out-of-scope items with the reason for each. Feature: [name it]. Save it to docs/specs/[slug]/prd.md.
to-issues (PRD → issues) — Sonnet 5 · branch plan/<feature>
Slice the PRD into small, visible, vertical tasks.
Use the to-issues skill. Break docs/specs/[slug]/prd.md into small issues sliced VERTICALLY (each touches schema + service + a visible bit of UI). Add blocking relationships, label each human-in-the-loop or AFK, keep traceability to the PRD. Write them to docs/specs/[slug]/issues.md.
tdd — Sonnet 5 · branch feature/<slice>
Build one slice at a time, test-first.
Use the tdd skill to implement: [the issue or feature]. Go red → green → refactor: write a failing test first and show it fail, then the minimum code to pass, run the suite, then refactor while tests stay green. Run tests AND type-check every cycle and report the real result — never say done without a passing run.
security-review — Opus 4.8 · current branch
Audit a change before you keep it: checks the OWASP Top 10, rates every finding by severity, and only reports — it doesn't touch the code.
Review the changes on the current branch for security. Check the OWASP Top 10 — injection, broken auth, hard-coded secrets, missing rate limiting, weak transport. Rate each finding Critical / High / Medium / Low and give a concrete fix with a code example. Don't change code — just report.
frontend-design (official) — Sonnet 5 · branch feature/ui
A production-grade interface, no generic AI look.
Use the frontend-design skill. Build [describe the page or component] as a production-grade interface: real typographic hierarchy, considered spacing, a deliberate palette, accessible, responsive, light and dark themes. Use [your stack]. Then show it running.
pptx (official) — Sonnet 5
A design-forward deck you can download.
Use the pptx skill to create a [N]-slide PowerPoint about [topic] for [audience]. One idea per slide, design-forward styling, a title slide and a closing call-to-action. Deliver the .pptx file to download.
xlsx (official) — Sonnet 5
An Excel model with real formulas.
Use the xlsx skill to build an Excel model for [purpose] with these inputs: [list them]. Add formulas, clear headers, sensible number formatting, and a summary sheet with the key outputs. Deliver the .xlsx file.
A full feature in one sitting: grill-me → to-prd → to-issues → tdd (per slice) → security-review.
How to improve any prompt
Three tweaks that raise the quality of almost any instruction:
- Ask for the reasoning, not just the result. "Explain why" pulls deeper answers.
- Give it a role. "Act as a corporate lawyer with 20 years of experience" changes the level of the answer.
- Never accept the first result as final. Iterate: "this is good, but change X". The best comes on the second or third pass.
That closes the fundamentals. In the next lesson: Handoff — how to continue a long piece of work in a fresh session without losing the thread. And in the modules that follow, we connect Claude to your real tools — Legal, Marketing, Web Design — and apply all of this to your profession.