Skills that do engineering work,
not demos.

Skills

They compose, and none requires the others: ticket → impact-map → proof-driven-dev → production-guard → ship  —  then practical-localizer when the product goes to another language, and incident → engineering-investigator → cause when something is already broken. project-compass runs underneath all of it, and asks whether the ticket is the right next thing at all; standards-compass asks what the software should have been measured against in the first place. dependency-guard and api-contract-guard step in at the two moments a change becomes hard to take back: a dependency arriving, and an interface someone else will build against. And deployment-compatibility answers the question that only exists once you name a server: does this project fit that machine? And architecture-engineer answers the one that comes before any of them — what shape should this system have, and why — but only when you ask it. skills-pipeline runs whichever of them you pick, in your order — also only when you ask.

Automatic activation

There are no slash commands to remember. The agent reads each installed skill’s description — which says when it applies and when it does not — and brings in the ones a request needs: several at once when several apply, none when none do.

⚡ Impact Map · Standards Compass — rename reaches report SQL; export carries personal data

Measured, not asserted: an activation suite runs real sessions against fixture repositories and checks which skills load — and, just as often, which must not. Naming a skill still works, and so does telling one to stand down. How they work together →

Why these exist

Coding agents are good at writing code and bad at knowing what the code touches. The expensive failures are not syntax errors — they are the raw SQL nobody grepped for, the nightly job comparing a status string, the fixture that keeps CI green while the behavior is wrong.

These skills exist to close that gap: they encode the investigation an experienced engineer does before touching anything, so the agent does it too, every time, and shows its evidence.

Installing

Skills install through the Skills CLI and work with any Agent Skills-compatible agent.

All twelve skills, any compatible agent

--skill '*' selects every skill in this repository, and the CLI asks whether to install them into the current project or globally. Add -g -y to skip the prompts and install globally, or -y alone to install into the current project. With no -a, the CLI installs to the agents it finds on your machine. Prefer this over the CLI's --all, which writes into every agent the CLI supports, not just the ones you have.

Claude Code only

--copy keeps the skills out of ~/.agents/skills, a folder Codex, Cursor and other agents also read.

Or as a plugin, which keeps the twelve together and updates them in one step:

Install one way or the other, not both, or each skill shows up twice.

One skill

Installed skills are selected by their description, so there is no slash command to memorize — work normally and the agent brings in what the request needs. In Claude Code, also add the ten-line standing instruction — it is what gets Claude to reach for the skills during implementation work — and, optionally, a colored active-skills status line: integrations/claude-code.

Uninstalling

Uninstalling should be as easy as installing. It is your right. One command removes every Skills Hub skill from every agent it went into, and nothing else. Your other skills stay exactly where they are.

Any compatible agent

skills-hub stands for all twelve skills. The command removes each one by name through the Skills CLI, and never uses --skill '*', which would remove every skill on the machine. Skills that are not installed are skipped. Add --dry-run to see exactly what it would remove, or --project to remove a project-level install instead of a global one.

Claude Code plugin

Nothing is left running. What stays is yours: lessons in ~/.skills-hub/ (rm -rf ~/.skills-hub), and project notes such as .project-compass/ in your repositories, which you can delete or keep. Full list.

What is an Agent Skill?

A folder containing a SKILL.md file: frontmatter naming the skill and describing when it applies, plus markdown instructions the agent follows when it does. Supporting reference files load on demand, so deep guidance costs nothing until it is needed.

There is no runtime and no engine. The intelligence is the agent's; the skill supplies the methodology — which is exactly why a good one is worth writing carefully.