Skip to content

jouet setup ​

  • Usage: jouet setup [FLAGS] <HARNESS>

Install jouet's hooks into a coding agent's harness

Merges hooks into Claude Code's settings.json, Codex's hooks.json, Cursor's hooks.json, Gemini CLI's settings.json or Copilot's hooks/jouet.json, keeping every other hook. The session-start hook primes each session with the commands and rules, so the skill, which says the same, is installed only with --skill or when priming is off ([hooks] prime). Running it again changes nothing. The jouet-init project onboarding skill is installed only with --init-skill; --uninstall removes only what jouet added. The hooks do nothing where jouet is not on PATH.

For Claude Code, setup also offers a context budget from your own recent sessions: a compaction window when they run on a 1M-token model, and denying what none of them used. Asked at a terminal, otherwise only noted: nothing is written without --compact-window or --deny-unused, and --undo takes each back.

Arguments ​

  • <HARNESS> — claude, codex, cursor, gemini or copilot

Flags ​

  • --global — Install for every project, under your home directory (the default)

  • --project — Install into this repository for you alone: Claude Code's untracked .claude/settings.local.json

  • --shared — Install into the repository's tracked hooks file (.claude/settings.json, .codex/hooks.json, .cursor/hooks.json, .gemini/settings.json or .github/hooks/jouet.json), for everyone who clones it (implies --project)

  • --dry-run — Show what would change without writing anything

  • --uninstall — Remove jouet's hooks and skills

  • --skill — Install the skill as well, which the session-start primer otherwise stands in for

  • --init-skill — Install the jouet-init skill, with which an agent fits jouet to a project's checks

  • --benchmark — Print a prompt for an agent to benchmark jouet on this project

  • --compact-window <TOKENS> — Claude Code: compact a conversation once its context reaches TOKENS (100k to 1m), by setting autoCompactWindow. It pays on a 1M-token model; a 200k one already compacts there

  • --deny-unused — Claude Code: deny the tools and MCP servers that none of your last 50 sessions called, and hide from the model the skills none used (still there as /name)

  • --undo <CHOICE> — Claude Code: take back what --compact-window (window) or --deny-unused (unused) added, and nothing else

    Choices: window, unused

  • -h --help — Print help

MIT LicenseCopyright © 2026jdx.dev