Install into tools

incantory add writes a prompt to the place each coding tool reads it from, converting the format where needed, and records what it installed in incantory.lock.

npx incantory add alice/code-review@production --to claude
npx incantory add alice/style-rules --to cursor
npx incantory add alice/pdf-skill@v3 --to claude --global

incantory add <owner/slug[@label|@vN]> --to claude|cursor|copilot|agents|codex|windsurf|gemini [--project|--global] [--yes-i-read-it] [--var name=value]...

--project (the default) writes into the current directory; --global writes into your home directory where the tool supports it. Every prompt page also has an Install box with the command for each tool.

Where files go

Which file a prompt becomes depends on its kind: a skill installs as a skill folder, an agent as an agent file, rules as a section of the tool's instructions file. Conversions that lose something (a skill's scripts cannot live in a .mdc rule, for example) print a warning.

Managed sections

When a prompt goes into a shared file such as AGENTS.md or CLAUDE.md, it is written between markers:

<!-- incantory:alice/style-rules@v4 start -->
...
<!-- incantory:alice/style-rules@v4 end -->

Sections are matched by owner/slug, so running add again replaces the section instead of duplicating it, update rewrites it in place, and remove cuts it out. Anything you wrote outside the markers is never touched.

incantory.lock

Installs are recorded under each entry's installs in ./incantory.lock (or ~/.config/incantory/incantory.lock for --global), next to the pinned version and content hash. Commit the project lockfile so everyone on the team gets the same versions.

npx incantory update                 # re-resolve labels and latest refs, rewrite installs
npx incantory update alice/style-rules
npx incantory verify                 # check pins against the registry
npx incantory remove alice/style-rules --to cursor

update only moves refs that follow a label or the latest version; @vN pins never change.

The risk gate

Skills, rules and agents can instruct a tool to run commands, so every package is scanned and carries a risk level. add and update show it before writing anything. A package rated high risk is refused unless you pass --yes-i-read-it, after reading what it does (exit code 6 otherwise).

Without the CLI

The same conversions are available as downloads: GET /api/v1/prompts/{owner}/{slug}/download?format=... with skill, cursor-mdc, copilot-instructions, copilot-agent, claude-agent, agents-md, claude-md, gemini-md, plain, zip or tar. The MCP tool incantory_convert returns the files for a model to write.