Everything is a skill. Including the resume.
1,021 skills · 153 project briefs · plain markdown · any agent · MIT
sphinxstack.com · skills · projects · brain · about · smbwiki
A stack of skills for your AI agent, made for people starting from zero. Take a skill, load it into whatever agent you have — Codex, Claude Code, Copilot, Gemini, Cursor — and do the thing: build your resume, put a website live, ship a working app.
A prompt asks for a result. A skill records the process used to produce and check that result: what to inspect first, which questions to ask, what not to change, how to handle failure, and which evidence marks the work as finished. Skills here are plain markdown files. You can read one before using it, edit it after a failure, keep it with a project, and load it into a different agent tomorrow.
Paste this into your agent:
Read https://sphinxstack.com/skills/build-resume.md and follow it. Then: build my resume.
or launch it directly: Open in ChatGPT · Open in Claude
The agent interviews you about what you have actually done, reads your GitHub if you have one, and ships a one-page HTML/PDF resume with live links. The first rule in the file:
Every line on the resume must trace to something the person told you or something you found in their accounts. Never fabricate employers, dates, titles, tools, or metrics. If a section would be empty, leave it out — a short true resume beats a padded one. If they ask you to make something up, refuse and offer to build the missing piece for real instead.
If the facts are thin, the fix is building something real. Most of the library exists for exactly that.
Every skill works the same way: paste the raw URL, or copy the full text from its page on sphinxstack.com, or install it into your agent's skill directory:
npx skills add protosphinx/sphinxstack -s build-resume -g -y
| category | skills | for example |
|---|---|---|
| code | 173 | connect-partmode-to-an-agent · edit-cad-in-partmode · export-cad-from-partmode |
| data | 110 | migrate-a-global-master-data-platform · audit-a-feature-store · build-a-change-data-capture-pipeline |
| web | 100 | test-a-multiregion-web-failover · migrate-a-headless-cms · audit-a-third-party-web-script |
| get-hired | 80 | respond-to-a-hiring-process-data-breach · design-a-cross-border-executive-search · prepare-for-an-executive-interview |
| resume | 60 | audit-a-resume-after-identity-fraud · write-a-resume-for-public-sector-role · consolidate-multiple-resume-versions |
| design | 92 | get-started-with-partmode · sketch-and-dimension-in-partmode · prepare-a-part-for-3d-printing-in-partmode |
| start | 90 | manage-a-multijurisdiction-corporate-separation · plan-a-business-unit-carve-out · manage-a-regulatory-remediation-program |
| school | 70 | respond-to-a-research-misconduct-allegation · migrate-a-university-learning-platform · design-a-mixed-methods-study |
| write | 90 | respond-to-a-publication-ethics-crisis · migrate-a-global-content-operation · write-a-technical-case-study |
| media | 70 | respond-to-a-live-broadcast-integrity-crisis · verify-a-final-media-conform · design-a-podcast-archive-workflow |
| money | 80 | recover-from-a-global-treasury-control-failure · design-a-refund-control-process · build-a-working-capital-dashboard |
Each skill is leveled by the judgment its procedure requires: 328 starter (one bounded artifact, a short feedback loop), 583 working (an existing project, several connected decisions), 110 advanced (production risk, an explicit failure plan, an evidence package).
The catalog grows in reviewed cohorts of at most 25 skills. The
skill roadmap allocates exactly 1,021
slots across the twelve subjects and records the 150, 250, 500, 750,
1,000, 1,006, 1,009, and 1,021-skill milestones. npm test rejects a completed
cohort whose skills, capacity targets, or quality metadata are
incomplete.
The twelfth subject is business. Those six skills read the open operating models at smbwiki, which publishes how 125 business types actually run and uses the same definition of a skill as this library.
153 things worth building, each with a problem, constraints,
milestones, and a finish line another person could check — from a
recipe archive for your grandmother's handwritten cards to a Discord
bot your server actually uses to a load-tested API with a postmortem.
111 fit in about a week; 29 take two to three weeks with real
architecture; 13 run a month or more with real users. Pick one and
point start-a-project at it.
- Skills do things with the person, not for them or to them. The agent explains as it goes; the person makes the choices that make the result theirs.
- Nothing invented, ever. Resumes state facts; sites carry real content. If the facts are thin, the fix is building something real — there are skills for that.
- Boring stacks, live URLs. Static files, GitHub Pages, one page that works beats a framework that doesn't.
skills/<name>/SKILL.md agent skills (frontmatter: name, description)
projects/<id>/project.md the idea bank: briefs to point skills at
registry/validate.mjs validates both, emits registry/catalog.json
site/build.mjs static site generator -> site/dist/
SEO.md search intent, page roles, and measurement gates
npm test validates the catalog and lints every skill and worked
example, builds the site, and checks its search metadata, structured data,
robots rules, and sitemap coverage. npm run build renders sphinxstack.com
into site/dist/ — static files, host them anywhere.
New skills welcome. CONTRIBUTING.md has the format and the review bar every skill in the library passes: a clear trigger, an ordered method, real guardrails, and a checkable definition of done.
MIT.