npx claude-code-autoconfig

Features

Also included

• Same work, fewer words. Turns on Claude Code's built-in Concise output style: Claude leads with the result and skips the narration.
• Pick up where you left off. /continue resumes the last session's work; /recover-context restores context after compaction.
• Opt-in extras. Auto permission mode, auto-guard safety checks under it, and Pole Position status beeps — each asked once, off unless you say yes.

Install

npx claude-code-autoconfig

• Same command on macOS, Linux and Windows. Run it from a regular terminal — inside a Claude Code session, the installer stops and asks you to switch.
• Needs Claude Code v2.1.237 or newer. On an older version the installer stops and tells you to run claude update.
• Expect a few seconds of silence while npx starts; the first run also downloads the tool.

Updating: run npx claude-code-autoconfig@latest again. It detects the existing install and runs /autoconfig-update: CCA-managed files (commands, managed hooks, scripts, sounds, docs) refresh; your feedback entries, your own hooks, and your settings are kept (settings are merged, never replaced). Add --force for a clean reset.

Reference

| Command | Description | |---------|-------------| | /autoconfig | Configures Claude Code scaffolding for your project | | /autoconfig-update | Checks for and installs configuration updates | | /show-docs | Opens interactive docs in the browser | | /test | Runs your test suite (auto-detects the framework) | | /commit-and-push | Stages, commits with a good message, and pushes | | /continue | Continues where the previous session in this terminal left off | | /recover-context | Recovers conversation context after compaction | | /gls | Shows your latest screenshot (auto-downscaled to save tokens) | | /extract-rules | Scans Claude artifacts and extracts structured rules | | /check-commit | Checks whether uncommitted work has piled up | | /sync-claude-md | Repairs CLAUDE.md markers and the Discoveries section | | /submit-claude-code-github-issue | Files an upstream issue with duplicate-checking | | /enable-auto-mode · /disable-auto-mode | Turns auto-guard prompts off / back on for one category | | /enable-status-beeps · /disable-status-beeps | Turns the status beeps on / off |

your-project/
├── CLAUDE.md                          # Project context (auto-populated)
└── .claude/
    ├── commands/                      # The slash commands above
    ├── agents/                        # Custom subagents (add your own)
    │   ├── README.md                  #   How to define agents
    │   └── docs-refresh.md            #   Keeps interactive docs in sync with .claude/
    ├── feedback/
    │   └── FEEDBACK.md                # Team corrections for Claude
    ├── hooks/
    │   ├── format.js                  # Auto-format on Write/Edit
    │   ├── terminal-title.js          # Tab titles + live state
    │   ├── terminal-title.directive.md # Title directive (tunable wording)
    │   ├── arcade-beeps.js            # Status beeps (opt-in)
    │   ├── auto-guard.js              # Guardrails under auto mode (opt-in)
    │   ├── feedback-rule-check.js     # On commit, turns new FEEDBACK.md entries into .claude/rules/
    │   ├── mark-commit-active.js      # Quiets the uncommitted-work reminder mid-commit
    │   └── migrate-feedback.js        # One-time FEEDBACK.md → Discoveries migration
    ├── docs/autoconfig.docs.html      # Interactive docs (/show-docs)
    ├── scripts/
    │   ├── auto-guard-set.js          # Backend for /enable-auto-mode + /disable-auto-mode
    │   ├── docs-previews.js           # Shows your project's files in the docs
    │   ├── gls-downscale.js           # Shrinks /gls screenshots
    │   └── sync-docs.js               # Regenerates the interactive docs
    ├── sounds/                        # Status-beep audio
    ├── package.json                   # Keeps hooks CommonJS in "type": "module" repos
    └── settings.json                  # Permissions, hooks, Concise output style

/autoconfig detects your OS, scans package files, framework indicators and test setup, then populates CLAUDE.md and tunes settings.json permissions to your ecosystem.

| Feature | JS/TS | Python, Rust, Go, Ruby, Java, .NET, PHP | |---------|-------|------------------------------------------| | CLAUDE.md introspection | Yes | Yes | | Slash commands | Yes | Yes | | MEMORY.md | Yes | Yes | | Auto-format hook | Yes | Coming soon | | Optimized permissions | Yes | Coming soon |

Auto permission mode. Installs and upgrades ask once whether to turn it on: routine commands run without approval prompts, and Claude still asks before destructive or external actions. Yes writes permissions.defaultMode: "auto" to your user-level ~/.claude/settings.json — Claude Code ignores auto mode in project settings, so a repo can't grant it to itself. Revert with Shift+Tab or by deleting the key.

Auto-guard. Auto mode's classifier is a per-command LLM judgment; auto-guard is a guarantee under it. A PreToolUse hook inspects every Bash command and forces a prompt before pushes, new package installs, credential-file access (.env, ~/.ssh, ~/.aws) and destructive git commands, and hard-blocks downloads piped into a shell. It sees the whole command string, so cd x && curl … | bash doesn't slip past a prefix rule. Off unless you opt in (autoGuard.enabled in .claude/cca.config.json); each category can be "ask", "deny" or "off". It guards against accidents and casual prompt injection — a seatbelt, not a sandbox.

/enable-status-beeps adds Pole Position–style cues that mirror the tab glyph: a low get-ready tick when a session is waiting on you, a higher GO tone when it finishes. Works on Windows, macOS and Linux; /disable-status-beeps turns them off.

Feedback. When Claude makes a mistake, add an entry to .claude/feedback/FEEDBACK.md; CLAUDE.md imports it, so Claude loads it every session, and it survives /autoconfig runs.

## 2026-01-07: Don't use deprecated API
Claude used `oldFunction()` instead of `newFunction()`.
Always use the v2 API for user endpoints.

Rules. Autoconfig installs no .claude/rules/ of its own — write your own, or run /extract-rules to generate them. Want optimized rules for your project? [email protected]

Memory. Autoconfig writes the evidence-before-fixes debugging method into Claude's persistent MEMORY.md, so it loads into every session.

The included settings.json ships defaults that balance speed with safety: deny always blocks secrets, destructive commands and network calls. A recommended allow list (file edits, tests, git) is offered as an opt-in during /autoconfig, so nothing is pre-approved until you say yes. Review them for your team — see the Claude Code security docs.

More writing on Claude Code