Skip to content
Docs: Installing

In your repos

Installing in a repository

One command writes a repo’s skills, its AGENTS.md block and a small CLI. Commit them, and every clone has the skills.

On this page

The first install

Open the repo in Skillify, go to Install and copy the line. In the root of a checkout:

shshell
curl -fsSL https://skillify.yogan.dev/i/<token> | sh

The link works for one hour and only for that repo, and it can only read. The script needs Node.js 20 or newer; it downloads the CLI and runs install, which writes the files below. Then commit them:

shshell
git add -A && git commit -m "Add agent skills"

Agents can fetch the same command themselves with the MCP tool get_install_command.

What gets written

Everything is plain files, meant to be committed. A fresh clone has its skills with no sign-in, no network and no install step, which is what makes them show up for everyone.

Path, What it is
PathWhat it is
.agents/skills/<name>/Each skill, overlay included. Read by Codex, Cursor, Copilot and other agents.
.claude/skills/<name>/The same skills for Claude Code.
AGENTS.mdA managed block between <!-- skillify:start --> and <!-- skillify:end --> listing the skills and the repo’s notes. The rest of the file is never touched.
CLAUDE.mdCreated with @AGENTS.md if the repo has none, so Claude Code reads the same block. An existing one is left alone.
.skillify/config.jsonThe server, the repo name, its ingest key and usage (on or off).
.skillify/lock.jsonThe version and hash of every skill written, so edits in the repo can be spotted.
.skillify/cli.mjsThe CLI: one file, no dependencies.
.claude/settings.jsonThe usage hook (only with usage on). Other settings in the file are kept.

Targets

By default skills are written to both .agents/skills and .claude/skills. They are the same files: one copy for agents that read the open .agents folder, one for Claude Code. If nobody on the repo uses one of them, turn that target off in the repo's Settings; the next sync removes it.

Keeping it current

When the library changes, the repo shows as behind. Anyone with a token runs:

shshell
node .skillify/cli.mjs sync
git commit -am "Update agent skills"

Sync needs a personal token: set SKILLIFY_TOKEN, or run node .skillify/cli.mjs login once. Get a token from Settings → API tokens; a read token is enough.

Local edits

Sync never overwrites a skill someone edited in the repo. It stops, names the skill, and gives two ways out: turn the edit into a proposal so every repo gets it, or run sync --force to drop it. To tailor a skill for this repo only, use an overlay instead of editing the files.

A check for CI

status --check exits with 1 when a skill was edited in place or is missing, and, with a token, when the repo is behind. It also reports the repo's state to Skillify, which keeps its drift fresh.

.github/workflows/skills.yml
name: Agent skills
on: [pull_request]
jobs:
  skills:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with: { node-version: 22 }
      - run: node .skillify/cli.mjs status --check
        env:
          SKILLIFY_TOKEN: ${{ secrets.SKILLIFY_TOKEN }}

Without SKILLIFY_TOKEN the check still catches local edits; with a read token it also fails when the repo is behind.

Terminalshell
node .skillify/cli.mjs status --check

Exit code 0 when current, 1 otherwise. Add --json for machine-readable output.

With a personal token you can set up any registered repo from its root, no install link needed:

shshell
curl -fsSL https://skillify.yogan.dev/cli.mjs -o /tmp/skillify.mjs
SKILLIFY_TOKEN=sk_… node /tmp/skillify.mjs init acme/payments-api --server https://skillify.yogan.dev