How it works

One command.
Then she's just there.

No account to create. No config to write. No settings to learn.

Install Claudia

Open your terminal and paste this:

$ npm install -g claudia-mentor

That's it. The installer automatically configures your shell. Restart your terminal, type claude, and Claudia is there. Every session. Forever.

To verify she's loaded, type /claudia:ask are you there? inside Claude Code.

Three modes

She shows up three different ways

Automatic checks

Every time Claude writes or edits a file, Claudia scans it. Leaked API key? She blocks it before it lands. Dangerous pattern like eval()? She warns you. You don't do anything — she just watches.

Joins the conversation

When you ask Claude something like "should I use MongoDB or Postgres?" or "how should I set up auth?", Claudia recognizes it as a technology decision and jumps in with opinionated advice. You don't summon her — she just knows when to show up.

Ask directly

Type /claudia:ask followed by any question:

/claudia:ask what database should I use for time-series data?
/claudia:explain (explains the code that was just written)
/claudia:review (reviews your recent changes for bugs)
/claudia:why (explains why your project uses its tech stack)
/claudia:health (runs a full project audit)
/claudia:wtf (breaks down an error: what / why / how to fix)
/claudia:where (guided tour of your project structure)
/claudia:resume (picks up where you left off)
/claudia:setup (first-time onboarding walkthrough)
/claudia:start (create a new project from scratch)
/claudia:shortcuts (keyboard shortcut reference)

Knowledge

What she knows

10 topics. Real opinions, not "it depends."

Databases

"Should I use Postgres or Mongo?" She'll tell you.

Security

Auth patterns, secrets management, common mistakes.

Infrastructure

AWS vs GCP vs Vercel. Containers. Serverless. Edge.

Frontend

React vs Svelte vs Vue. SSR vs SPA. State management.

API Design

REST vs GraphQL vs tRPC. Versioning. Real-time.

Testing

What to test. What not to test. When to mock.

Performance

N+1 queries. Bundle size. Core Web Vitals.

DevOps

CI/CD pipelines. Deployments. Monitoring.

Data Modeling

Schema design. Migrations. Indexing strategy.

Prompt Coaching

Detects vague prompts. Suggests better ones.

Automatic checks

What she catches

8 checks that run automatically on every file write. Plus 7 proactive hooks that watch your conversation.

blocks

Leaked secrets

AWS keys, API tokens, passwords, private keys, connection strings. Blocked before they touch the file.

warns

Bad patterns

eval(), empty catch blocks, console.log in production, SQL concatenation, HTTP URLs, chmod 777.

warns

Sketchy dependencies

Deprecated packages, compromised packages, trivial packages that should be inline code.

warns

Docker issues

Running as root, huge base images, secrets in ENV, missing multi-stage builds.

blocks

Git hygiene

Blocks .env file writes and merge conflict markers. Warns on large binaries.

warns

Accessibility

Missing alt text, unlabeled inputs, icon-only buttons, click handlers on divs.

warns

License compliance

Catches GPL/AGPL dependencies sneaking into your MIT-licensed project.

warns

CSS anti-patterns

!important abuse, hardcoded colors, extreme z-index values, float-based layouts.

Beginner mode

Never coded before? She adapts.

Set experience: "beginner" in your context file and Claudia changes how she shows up.

~/.claude/claudia-context.json
{
  "experience": "beginner"
}

Simplified greeting

No command list on startup. Just "Claudia is here. Just build. I'm watching." Commands get revealed when they're relevant.

Stuck detection

Type "I'm stuck", "help", "idk", or "where do I start" and Claudia asks one clarifying question, then suggests one small next step.

Run suggestions

After Claude creates a file, Claudia tells you how to run it: open index.html, python3 app.py, npm install.

Next-step suggestions

When Claude finishes a task, Claudia suggests 2-3 contextual things you could try next. No dead ends.

Milestones

Celebrates real achievements: first file created, first bug fixed, first commit, first project running, 10+ files. Persists across sessions.

Progressive command reveal

Commands appear only when relevant. Error? She mentions /claudia:wtf. File created? /claudia:explain. Growing project? /claudia:where.

Customization

Dial her up or down

Create ~/.claude/claudia.json to change how she behaves.

~/.claude/claudia.json
{
  "proactivity": "high"
}
"low"

Only responds when you use /claudia. Quiet mode.

"moderate"

Jumps in on security issues and major decisions. This is the default.

"high"

Flags everything. Full mentor mode. Great for learning.

Want to see everything?

Every knowledge file and hook script is open source on GitHub.

Browse the source