promptheus

Interactive installer for the Promptheus toolkit — one command, three pickers, a repo your AI agent actually knows how to work in.

npx promptheus init

It scans your repo, detects your stack and your agent tool, then walks you through three full-catalog multi-select screens (everything relevant is pre-checked — untick what you don't want):

1. Rules — 50+ production-grade rule sets (Next.js, Django, Go, Rust, Docker, TDD, Security…). Every rule you check is merged into one AGENTS.md — the open standard read by Claude Code, Codex, Cursor, Gemini and 30+ agents. In a monorepo each app gets its own AGENTS.md with its stack rules, plus a root file with the shared conventions and an app index.
2. Subagents — a specialist crew (code reviewer, debugger, security auditor…) installed to .claude/agents/ (Claude Code only).
3. Slash commands — /review, /commit, /test, /debug… installed in your tool's native format and directory.

Then it wires the toolkit together: an ## Installed AI toolkit section is appended to your root AGENTS.md listing every installed subagent and slash command with its trigger, so your agent delegates and uses them proactively — without being asked.

Options

| Flag | Description | | ---- | ----------- | | -y, --yes | Accept the detected defaults (skip the pickers) | | --all | Select the entire catalogs (skips the pickers too) | | -f, --force | Overwrite files that already exist | | --dry-run | Show the full plan — files, merge composition, wiring — without writing | | --tool <t> | Target claude | cursor | gemini | codex | windsurf | | --dir <path> | Run against another directory | | -v, --version | Print the version |

--cursor, --gemini, --codex and --windsurf are shorthands for --tool.

Non-interactive usage

npx promptheus init --yes          # the detected default kit, no questions
npx promptheus init --all --force  # everything, overwriting existing files
npx promptheus init --dry-run      # print the plan, write nothing
npx promptheus list                # print the grouped catalogs
npx promptheus list rules          # …or just one of rules | agents | commands

Without a TTY (CI, scripts), init behaves like --yes.

Where things go

• AGENTS.md — the merged rules + the toolkit wiring section (always, whatever the tool — it's the open standard)
• .claude/agents/*.md — subagents (Claude Code)
• .claude/commands/, .cursor/commands/, .gemini/commands/ (TOML), .codex/prompts/, .windsurf/workflows/ — slash commands per tool
• README.md — an optional agent-ready badge (one line, idempotent)

Already-existing files are never touched unless you pass --force.

Env overrides

Everything is fetched live from the Promptheus sites; point the CLI elsewhere (mirror, staging, local dev) with:

| Variable | Default | | -------- | ------- | | PROMPTHEUS_RULES_URL | https://rules.promptheus.dev | | PROMPTHEUS_AGENTS_URL | https://agents.promptheus.dev | | PROMPTHEUS_COMMANDS_URL | https://commands.promptheus.dev | | PROMPTHEUS_HUB_URL | https://promptheus.dev |

If the rules index is unreachable, the CLI falls back to an embedded snapshot of the catalog so init keeps working.

Zero dependencies, copy-paste, public domain (CC0). Nothing is uploaded. Part of promptheus.dev.