Reference

CLI

Every command and flag of the prongs CLI. Generated from the same data as prongs --help.

Run it with npx prongs, or through your package manager’s scripts. It needs Node 22.18 or newer: it loads your TypeScript registry with Node’s built-in type stripping.

Commands

`prongs add`

Copy the UI components and starter intents into your repo, and write prongs.json.

Prop
What it does
--dir
Where the components go. Defaults to components/prongs.
<path>
--force
Overwrite files that already exist.
flag

Terminal

npx prongs add

`prongs add gallery`

Add a page that renders every intent in your registry live (development only by default).

Prop
What it does
--route
Where to put the page in a Next.js app. Defaults to app/prongs/page.tsx.
<file>
--public
Serve the page in production too.
flag

`prongs add logs`

Add the logs profile: audit, security, incident, job and debug intents, plus log().

`prongs add errors`

Add the errors profile: request, auth, permission, record, conflict, limit and failure intents, plus fail(). Adds the logs profile too, which errors write to.

`prongs add analytics`

Add trackIntents(), which reports presses, submits, toggles and opens as events named by intent. You decide where they go.

`prongs find <what it is for>`

Look up the intents for a job: each match with what it renders, its "not for" lines and how code writes it.

Prop
What it does
--limit
How many matches to show (default 3).
<n>
--json
Print the matches as JSON, for tools.
flag

Terminal

npx prongs find "wrong email or password"

`prongs check [paths...]`

Check how intents are used. Existing violations in the baseline pass; new ones fail.

Prop
What it does
--json
Print the findings as JSON, for tools.
flag
--suggest
Group repeated overrides and styled raw elements into proposed intents.
flag
--update-baseline
Record today’s findings in .prongs/baseline.json, so only new ones fail.
flag
--no-baseline
Ignore the baseline and report everything.
flag
--max-unsafe
Fail when unsafeClassName is used more than n times.
<n>
--hook
Run as an agent hook: read the edited file from the hook payload and answer in that tool’s format.
claude | cursor

Terminal

npx prongs check --suggest

`prongs add-intent <id>`

Add an intent to the registry, validate it, and regenerate PRONGS.md. Rolls back if the registry would be invalid.

Prop
What it does
--description
Required: what the intent is for, in a sentence.
"<what it is for>"
--refs
Intents to build it from.
a,b
--classes
UI intents: its own classes.
"..."
--element
UI intents: the element it renders.
<tag>
--not-for
When not to use it.
"<context>"
--instead
What to use in that case.
<id>

Terminal

npx prongs add-intent action.follow --refs control.primary --description "Follows a person"

`prongs spec`

Write PRONGS.md from the registry.

Prop
What it does
--out
Where to write it. Defaults to PRONGS.md.
<file>
--agents
Also add a short block to AGENTS.md, and make sure CLAUDE.md reaches it.
flag
--check
Write nothing; fail if PRONGS.md is out of date. For CI.
flag

Terminal

npx prongs spec --agents

`prongs export`

Print the registry as JSON, in the shape the spec defines, for tools that do not run TypeScript.

Prop
What it does
--out
Write it to a file instead of printing it.
<file>

Terminal

npx prongs export --out prongs.registry.json

`prongs install-hooks`

Run prongs check on every file an agent edits.

Prop
What it does
--claude
Only the Claude Code PostToolUse hook.
flag
--cursor
Only the Cursor postToolUse hook.
flag

Common options

Prop
What it does
--config
Use a config file other than ./prongs.json.
<prongs.json>
--cli
How printed commands should call the CLI, e.g. "yarn prongs".
"<how to run prongs>"