A website, a database, a job search, the resume at the end. Each one can be a skill your agent knows how to follow.
sphinxstack is a working library of skills for Codex, Claude, Copilot, Gemini, and Cursor. Each skill gives an agent a method for one kind of work. Skills can refer to other skills; together, they form the skills layer of an agent's brain.
A skill tells the agent where to start, which steps to follow, which rules apply, and what must be true before the work is complete. New to the idea? What is a skill? explains it from zero.
Every skill enters the catalog in a reviewed cohort of at most 25, with an explicit finish and evidence another person can check.
The newest procedures added to the library, listed in publication order. Installs count successful loads of the raw SKILL.md file.
A skill is a reusable set of instructions that teaches an AI agent how to perform a specific kind of work. It records when the method applies, the order of work, the rules to follow, and the evidence required before the task is complete.
For example, the use-an-api skill tells the agent to read documentation before writing code, keep credentials out of version control, test failure cases, and verify the shipped feature.
A prompt describes a task or desired result. A skill supplies a repeatable
method for completing it. Connect this app to an API requests an
outcome; the skill specifies how the agent should produce and verify it.
Because a skill is plain text, you can inspect it, revise it after a failure, version it, and use it with another agent. Skills can also refer to other skills, allowing several focused methods to support one project.
These skills introduce processes, rules, and checks that can be reused across different projects.
Each file includes a trigger, procedure, guardrails, and definition of done. Skills are grouped by kind of work; choose the one that matches the task in front of you.
Turn real experience into documents and evidence for applications.
Find openings, apply, interview, and follow through.
Build systems that use data, APIs, automation, and accounts.
Build, publish, measure, and improve work on the web.
Turn an idea into a project, club, channel, or first sale.
Draft, edit, publish, and send writing in the person's own voice.
Make visual decisions, test them, and export usable artifacts.
Record, edit, caption, and publish audio and video.
Structure data, ask questions, and report what the evidence supports.
Handle first budgets, prices, invoices, and sales carefully.
Plan applications and study work without doing graded work for the student.
Document how a business runs, then hand parts of it to an agent.
A project brief defines the work on which skills can be used. Each brief has milestones, constraints, and a finish line sized to the depth of the build.
Finish one thing you can show.
Connect several parts into a working system.
Build for real users, then explain what you learned.
A project can use several skills, but it does not have to use all of them. Each skill owns one procedure, leaves behind evidence, and recommends another skill only when the project needs it. This is one possible sequence for taking a web app from an idea to a resume entry.
Choose the user, narrow the problem, and write a finish line that can be checked.
Create the smallest complete flow and make the core data persist.
Give each user an identity and test who can read or change every record.
Add outside data without exposing credentials, then test failed responses.
Check keyboard use, headings, labels, focus, contrast, and error messages.
Publish to a stable URL and record how to rebuild, inspect logs, and roll back.
Measure one useful question and verify that a real visit reaches the dashboard.
Turn the shipped work, decisions, failures, and proof into a defensible entry.
A project brief describes something worth building. It is not a tutorial. It supplies the shape of the project; skills supply the methods used to complete each part.
Skills are one of four layers in a useful agent setup. The other layers keep your methods attached to your preferences, your facts, and the checks that should happen every time.