Using Inyon
Agent skills
Skills that come with each release, like Blueprint for plan pages, and how to install them with inyon skill add or by hand.
A skill is a folder of instructions and files that a coding agent loads when a task calls for it. From 0.17.7, each Inyon release ships its skills next to the programs, signed the same way. Install them with one command, or download one and unpack it yourself.
Install with inyon
inyon skill list # the skills in the newest release, and where each is installed
inyon skill add blueprint
inyon skill remove blueprint
inyon skill add downloads inyon-skill-<name>.tar.gz from the newest release. It checks the file against the release's signed SHA256SUMS, as inyon update does, then unpacks it for each coding agent on the computer:
| Agent | Folder |
|---|---|
Claude Code (~/.claude exists) | ~/.claude/skills/<name> |
Codex (~/.codex or ~/.agents exists) | ~/.agents/skills/<name> |
Gemini CLI, Antigravity (~/.gemini exists) | ~/.gemini/config/skills/<name> |
If none of these folders exist, the skill goes in ~/.agents/skills. With --to <folder> it goes into that folder instead, for example a project's own .claude/skills. Agents pick the skill up in their next session.
Running add again replaces the copy with the newest one. It won't overwrite a skill of the same name that you put there yourself unless you add --force, and remove only deletes copies that inyon skill add installed.
Install by hand
Each skill is on the download page as inyon-skill-<name>.tar.gz, with its folder at the top level. Check it against the release's list as described in Checking a download. Then unpack it into your agent's skills folder:
tar -xzf inyon-skill-blueprint.tar.gz -C ~/.claude/skills
On Windows, tar comes with Windows 10 and 11: tar -xzf inyon-skill-blueprint.tar.gz -C $HOME\.claude\skills.
Blueprint
Blueprint turns a plan into a small website instead of a long chat message. Ask your agent for "a plan page", "an HTML plan" or "mockups of the redesign" and it builds a folder with four parts:
- The plan: the written plan, with a live "Where it stands" board on top and releases you can open one by one.
- Screens: clickable mockups in a viewer that switches between phone and desktop sizes and light and dark. Right-click, long-press, menus, sheets and toasts work.
- Feature matrix: one row per feature, with what exists today, its status and the release that brings it, plus search and filters.
- Progress board: read from
progress.json, which the agent updates while it works. The page checks it every 20 seconds.

The skill starts from a working sample site for a made-up grocery app, then replaces it with your project. Its scripts serve the site, take screenshots, and check every screen for console errors and buttons that do nothing:
| Script | What it does |
|---|---|
new-site.mjs <dir> --name "App" | Copies the template into a new folder |
serve.mjs <dir> --host <ip> --port 8790 | Serves the site. Use a private address, such as a VPN or home network IP, to open it on your phone |
stills.mjs <dir> --audit | Saves gallery pictures of every screen and reports errors and dead buttons |
shoot.mjs <url> <out.png> <w> <h> <light|dark> | One screenshot, with optional clicks before it |
progress.py task <id> doing|done | Updates the progress board |
The scripts need Node 18 or newer and Playwright with Chromium (bun add -d playwright, then bunx playwright install chromium). With sharp installed, the pictures are WebP. Without it, they're PNG.