Adaptive AI coding assistant configurator for Claude Code. Scans your project, detects the stack (95+ project types), selects the right hooks/MCPs/skills, and configures everything automatically.
npm i -g puiux-pilot
cd your-project
puiux-pilot init # dry-run: shows what it would do
puiux-pilot init --apply # apply changes (backup created first)puiux-pilot scan # project DNA without changes
puiux-pilot score # quality score (0-100, A-F)
puiux-pilot doctor # health check
puiux-pilot eject # clean removal| Command | What it does |
|---|---|
init |
Scan project, select hooks/MCPs/skills, show plan. Dry-run by default. |
init --apply |
Write changes (backup created first). |
init --force |
Overwrite user-modified hooks. |
scan [path] |
Analyze project without changing anything. Shows Project DNA. |
score [path] |
Quality assessment: 6 dimensions, 0-100, A-F grade. |
doctor |
Health check: broken hooks, missing scripts, timeouts, orphans. |
hooks list |
List all hooks with status. |
translate |
[EXPERIMENTAL] Translate config between AI coding tools. |
eject |
Clean removal. Restores original state from backup. |
- Dry-run by default —
initshows what would change without writing anything. - Backup before every apply — stored in
~/.puiux-pilot/backups/<timestamp>/. - Atomic writes — temp file + rename; no partial writes on crash.
- Rollback on failure — if any step fails, previous state is restored.
- Manifest-scoped uninstall —
ejectonly removes hooks Pilot installed. User hooks, permissions, and MCP configs are never touched. - No cloud dependency — everything runs locally.
Based on your project type, Pilot selects from:
- 28 hook scripts across 6 events (PreToolUse, PostToolUse, Stop, SessionStart, PreCompact, UserPromptSubmit)
- 7 profiles: full-stack, api-only, library, security-first, mobile, startup-speed, infra
- MCP auto-detection: ESLint, Prisma, Tailwind CSS, icons8, lottiefiles — based on actual dependencies
- Skills: clean-code, e2e-delivery, code-quality, design-standards, and more
Example: a Next.js webapp gets 28 hooks + 4 MCPs. A Node API gets 18 hooks + 0 MCPs.
GitHub Actions workflow runs on every push to main:
| Job | OS | What |
|---|---|---|
build-and-test |
ubuntu, windows, macos × node 18, 22 | Build + unit tests + CLI smoke |
release-gates |
ubuntu, macos | Full gates: 72 assertions (cold-start, repo matrix, custom hooks, failure injection) |
smoke-windows |
windows | Lightweight smoke: scan, init, apply, idempotency, doctor, eject |
Run gates locally: bash tools/release-gates.sh (uses sandboxed HOME, never touches real ~/.claude).
Cross-tool config translation between: CLAUDE.md, .cursorrules, .clinerules, .windsurfrules, .github/copilot-instructions.md, CONVENTIONS.md.
# Auto-detect source, generate all missing formats
node bin/puiux-pilot.mjs translate --auto
# Specific translation
node bin/puiux-pilot.mjs translate --from claude --to cursorThis feature is experimental. Round-trip fidelity is best-effort; tool-specific rules are preserved as comments.
PUIUX Pilot is free and open source for individual developers under the Apache 2.0 license.
Pro for teams (coming soon): policy packs, team sync, drift detection, private hook registry, and priority support.
See TRADEMARKS.md for brand usage guidelines.
Apache-2.0 — see LICENSE for details.
