roast

AI code reviewer that roasts your code with humor and honesty.

🚀 Launching on Product Hunt — April 9, 2026

👉 Support us on Product Hunt | Get notified when we launch!

What is this?

roast is a CLI tool that uses Claude to review your code and deliver feedback that's accurate, funny, and useful. Think Gordon Ramsay meets your senior developer.

Why use this?

Before:

You: "Is this code good?"
Brain: "Probably fine..."
*ships to production*
*everything breaks*

After:

$ roast bubble-sort.js

🔥 Oh boy, bubble sort in 2026? What's next, a floppy disk driver?

🔥 This function mutates the input array. That's like borrowing 
someone's car and returning it as a motorcycle.

💡 Use Array.prototype.toSorted() if you're on Node 20+

Real problems:

Code reviews take days
Teammates are too nice to be honest
You're working solo with no feedback
CI catches bugs after you commit
Writing tests doesn't catch design issues

roast gives you instant, brutally honest feedback before you commit. Use roast mode for your own code, serious mode for team reviews.

How is roast-cli Different?

Feature	roast-cli	ESLint	SonarQube	GitHub Copilot
Setup time	0 sec (npx)	Minutes	Hours	GitHub account
Personality	Gordon Ramsay 🔥	Dry/technical	Corporate	Helpful AI
Entertainment	✅ High	❌ None	❌ None	🤷 Varies
Code quality	✅ Real feedback	✅ Strict rules	✅ Deep analysis	✅ Suggestions
Use case	Quick laugh + insight	Enforce standards	Team quality	Code assistance
Cost	Free	Free	Paid tiers	$10/mo

Key Differentiator: We're the only tool that makes code review fun. Use us alongside your linter, not instead of it.

Installation

npm install -g @muin/roast

Or run without installing:

npx @muin/roast your-file.js

Setup

Get your Anthropic API key at console.anthropic.com

export ANTHROPIC_API_KEY="your-key-here"

Add to ~/.bashrc or ~/.zshrc to make it permanent.

Usage

Roast mode (default)

roast src/app.js

Output:

🔥 CODE ROAST 🔥
Victim: bubble-sort.js (JavaScript)
──────────────────────────────────────────────────

🔥 Oh boy, bubble sort in 2026? What's next, a floppy disk driver?

🔥 This function mutates the input array. That's like borrowing 
someone's car and returning it as a motorcycle.

💡 Use Array.prototype.toSorted() if you're on Node 20+, or 
at least clone the array first: const sorted = [...arr]

🔥 No input validation. Passing a string? Enjoy your runtime error.

✨ At least you got the algorithm right. It's bad, but it's 
correctly bad.

──────────────────────────────────────────────────
Roasted with ❤️  by Claude

Serious mode

roast --serious src/app.js

Output:

📋 Professional Code Review
File: api-handler.js (JavaScript)
──────────────────────────────────────────────────

🚨 No input sanitization on user data - SQL injection risk

⚠️  Synchronous file operations will block the event loop

💡 Consider using async/await with fs.promises

✅ Good error handling structure

──────────────────────────────────────────────────

Custom model

roast --model claude-opus-4 src/app.js

Supported Languages

JavaScript, TypeScript, Python, Go, Rust, Java, C, C++, Ruby, PHP, Swift, Kotlin, Shell, SQL, HTML, CSS, and more.

Options

-s, --serious              Professional review mode (no humor)
--severity <level>         Roast severity: mild, medium, harsh (default: medium)
-m, --model <model>        AI model to use (default: claude-sonnet-4-5)
--no-color                 Disable colors
-V, --version              Output version
-h, --help                 Display help

Severity Levels

Mild (😊 Be Nice Mode): Friendly, encouraging feedback. Perfect for beginners or when you want constructive criticism with a supportive tone.

roast --severity mild src/app.js

Medium (🔥 Default): Balanced mix of humor and criticism. Sarcastic but helpful, like a senior dev at code review.

roast src/app.js
# or explicitly:
roast --severity medium src/app.js

Harsh (💀 No Mercy): Brutally honest, savage roasts. Only use this if you can handle the truth. Gordon Ramsay mode.

roast --severity harsh src/app.js

Examples

Example 1: Quick file roast

$ roast src/utils/array-helpers.js

🔥 CODE ROAST 🔥
Victim: array-helpers.js (JavaScript)
──────────────────────────────────────────────────

🔥 You wrote a custom array flatten function? Array.flat() 
has been in JavaScript since 2019. ES2019 is not "too new."

🔥 uniqueArray uses indexOf in a loop - O(n²) complexity.
Set([...arr]) is O(n) and already built in.

💡 Half these functions are one-liners with modern JS:
   flatten: arr.flat()
   unique: [...new Set(arr)]
   last: arr.at(-1)

✨ At least they work correctly. But you've essentially 
reinvented lodash, poorly.

──────────────────────────────────────────────────

Example 2: Serious mode for team PR review

$ roast --serious src/api/auth.ts

📋 Professional Code Review
File: auth.ts (TypeScript)
──────────────────────────────────────────────────

🚨 Password comparison using === instead of timing-safe compare
   Risk: Timing attacks could leak password information

⚠️  JWT secret loaded from process.env without fallback check
   Will crash on startup if JWT_SECRET is not set

💡 Consider using express-validator for input sanitization

✅ Good: Proper async/await usage throughout
✅ Good: TypeScript types are well-defined

⚠️  Token expiration set to 30 days - consider shorter duration
   for sensitive operations

──────────────────────────────────────────────────

Example 3: Review from stdin (pipe or paste)

$ cat suspicious-code.py | roast

🔥 CODE ROAST 🔥
Victim: stdin (Python)
──────────────────────────────────────────────────

🔥 eval() on user input? That's not a security vulnerability,
that's a welcome mat for hackers.

🔥 Bare except: catches everything including KeyboardInterrupt.
You can't even Ctrl+C out of this disaster.

🔥 Global variables modified inside functions with no documentation.
Reading this code is like a mystery novel where the butler did it,
but also the gardener, and maybe the protagonist.

💡 Use ast.literal_eval() for safe evaluation, or better yet,
json.loads() if you're parsing data.

──────────────────────────────────────────────────

Example 4: Git diff review before commit

$ git diff src/payment-processor.js | roast --serious

📋 Professional Code Review
──────────────────────────────────────────────────

🚨 Changed error handling to swallow exceptions silently
   Original code logged errors, new code hides them

⚠️  Removed input validation for transaction amount
   Could now process negative or NaN values

🚨 API timeout increased from 5s to 60s
   May cause cascading failures under load

Recommendation: These changes reduce system reliability.
Suggest reverting the exception handling changes.

──────────────────────────────────────────────────

Example 5: Custom model for complex code

$ roast --model claude-opus-4 src/distributed-lock.go

🔥 CODE ROAST 🔥
Victim: distributed-lock.go (Go)
──────────────────────────────────────────────────

🔥 Your distributed lock implementation has a race condition
between checking and acquiring. Classic "check-then-act" bug.

🔥 Lock timeout is hardcoded to 10 seconds. Production load
spikes will turn this into a deadlock factory.

💡 Redis SETNX is atomic - use it directly instead of GET + SET.
Or better yet, use Redlock algorithm for multi-node safety.

🔥 Panic on Redis connection error. In distributed systems,
network failures are features, not exceptions.

✨ Good use of context for cancellation. That's the one part
that won't cause a 3 AM page.

──────────────────────────────────────────────────

Example 6: Mild severity for learning

$ roast --severity mild beginner-script.py

😊 Friendly Code Review
Victim: beginner-script.py (Python)
──────────────────────────────────────────────────

💡 Great start! Your code works, which is the most important part.

💛 Small improvement: Instead of opening files without 'with',
   try using context managers:
   
   with open('file.txt', 'r') as f:
       data = f.read()
   
   This automatically closes the file for you!

💡 You're using global variables. As your program grows,
   consider passing data as function parameters instead.

✨ Your variable names are clear and descriptive. Keep that up!

──────────────────────────────────────────────────

Example 7: Harsh mode for reality check

$ roast --severity harsh legacy-spaghetti.php

💀 BRUTAL CODE ROAST 💀
Victim: legacy-spaghetti.php (PHP)
──────────────────────────────────────────────────

💀 This code is a war crime. The Geneva Convention should cover this.

💀 mysql_* functions were deprecated 11 years ago. This isn't
   legacy code, it's a fossil.

💀 SQL injection vulnerabilities everywhere. You're concatenating
   user input into queries like it's 1999. It was bad in 1999 too.

💀 Password stored in plaintext. Just... why? WHY?!

💀 4 levels of nested if statements with no clear logic. This is
   a crime scene where the detective gave up and went home.

💀 Variable names like $a, $tmp, $data2. Were descriptive names
   too expensive?

Reality check: This needs a complete rewrite. I'm not even sure
where to start. Maybe with fire.

──────────────────────────────────────────────────
Roasted with absolute honesty by Claude

Example 8: TypeScript React component

$ roast src/components/UserDashboard.tsx

🔥 CODE ROAST 🔥
Victim: UserDashboard.tsx (TypeScript/React)
──────────────────────────────────────────────────

🔥 850 lines in one component. This isn't a component, it's a
   monolith. Break it down before it breaks you.

🔥 useEffect with 12 dependencies? That's not reactive programming,
   that's chaos with extra steps.

💡 Extract at least 4 sub-components:
   - UserHeader
   - ActivityFeed  
   - SettingsPanel
   - NotificationCenter

🔥 Inline styles everywhere instead of styled-components or CSS modules.
   Your future self will hate present you.

✨ Props are properly typed. That's good! Now use that discipline
   everywhere else.

──────────────────────────────────────────────────

Example 9: Shell script review

$ roast deploy.sh

🔥 CODE ROAST 🔥
Victim: deploy.sh (Shell)
──────────────────────────────────────────────────

🔥 No 'set -e' at the top. If step 1 fails, steps 2-10 will
   still run and cause chaos.

🔥 rm -rf without any confirmation? Living dangerously, I see.

💡 Add error handling:
   set -euo pipefail
   
   This will:
   - Exit on error (e)
   - Fail on undefined variables (u)
   - Catch errors in pipes (pipefail)

🔥 Hardcoded paths and credentials. Use environment variables:
   ${DB_HOST:-localhost}

✨ At least you have comments. That's rare for shell scripts.

──────────────────────────────────────────────────

Example 10: SQL query review

$ roast db/queries.sql

🔥 CODE ROAST 🔥
Victim: queries.sql (SQL)
──────────────────────────────────────────────────

🔥 SELECT * in production queries. Your DBA is crying somewhere.
   List the exact columns you need.

🔥 No indexes on the WHERE clause columns. This will be fast now
   with 100 rows. With 1,000,000 rows? Enjoy your 30-second queries.

💡 Add indexes:
   CREATE INDEX idx_users_email ON users(email);
   CREATE INDEX idx_orders_user_created ON orders(user_id, created_at);

🔥 LEFT JOIN without understanding why. If you need all rows from
   both tables, that's a FULL JOIN (or you don't actually need all rows).

✨ Query is at least readable and properly formatted.

──────────────────────────────────────────────────

Example 11: Rust code review

$ roast src/main.rs

🔥 CODE ROAST 🔥
Victim: main.rs (Rust)
──────────────────────────────────────────────────

🔥 .unwrap() everywhere. You learned Rust yesterday, didn't you?

💡 Use proper error handling:
   // Instead of:
   let file = File::open("data.txt").unwrap();
   
   // Use:
   let file = File::open("data.txt")?;
   // or
   let file = File::open("data.txt")
       .expect("Failed to open data.txt");

🔥 Cloning everything to avoid fighting the borrow checker.
   That's like buying a sports car and only driving in first gear.

✨ You're using match instead of if-let. Good! Pattern matching
   is the way.

💡 Learn lifetimes. They're scary at first, but they'll save you
   from these clones.

──────────────────────────────────────────────────

Real-World Scenarios

Pre-commit sanity check

# About to commit that refactor you've been working on
$ git diff HEAD src/auth.js | roast

🔥 You're exporting the private key in plaintext? Bold strategy.

🔥 This regex will match "admin@evil.com" as a valid admin email.
Better tighten that up unless you're running a very open company.

💡 hash.compare() is async but you're not awaiting it. This will
always return true. Always. Every time. 100% authentication success rate!

# ...okay maybe I should test this first

Reviewing that "quick fix" from 2am

$ roast src/hotfix-do-not-touch.js

🔥 File name is literally "do-not-touch.js" - that's a red flag
wrapped in another red flag

🔥 You're catching errors and logging "it broke lol". When production
is on fire, your logs will just say "it broke lol" repeated 50,000 times.

🔥 This setTimeout is set to 86400000ms. That's 24 hours. Hope nobody's
waiting for this response.

✨ The actual logic is... fine? But please, for the love of debugging,
add better error messages.

Checking tutorial code before copy-paste

$ curl -s https://example.com/tutorial.js | roast

🔥 This tutorial is using var in 2026. It was written during the
Mesozoic Era and hasn't been updated since.

⚠️  jQuery is loaded from an HTTP URL. That's a mixed content warning
waiting to happen.

💡 Modern replacement using fetch() would be 10 lines and zero
dependencies. Just saying.

The "I learned this yesterday" review

# Just picked up Rust, wrote first program
$ roast hello.rs --serious

📋 Professional Code Review
──────────────────────────────────────────────────

✅ Proper error handling with Result type

⚠️  .unwrap() on line 8 will panic on error. Consider using
    expect() with a meaningful message, or propagate with ?

💡 String ownership is correct, but you're cloning unnecessarily
   on line 12. Use a reference: &user_name

✅ Good use of match for control flow

Overall: Solid first program. Remove the unwrap() and you're good to go.

Legacy code archaeology

$ roast legacy/customer-import-final-v3-NEW-USE-THIS.php

🔥 Based on the filename, this has been "final" at least 3 times.
That's not a good sign.

🔥 mysql_connect() was deprecated in PHP 5.5 (2013) and removed in
PHP 7 (2015). This code is old enough to vote.

🔥 SQL query is concatenating user input directly. This is how
Little Bobby Tables drops your database.

🔥 No password hashing - passwords stored in plaintext. In the
event of a breach, this is "directly to jail, do not pass go" territory.

💡 Complete rewrite recommended. Start fresh with PDO and password_hash().
This is beyond roasting, this needs a Viking funeral.

Team code style check

# Check if the new junior's code matches your style
$ roast --serious src/components/UserCard.jsx

📋 Professional Code Review
──────────────────────────────────────────────────

⚠️  Component is 450 lines long. Consider splitting into smaller pieces.

⚠️  Inline styles instead of CSS modules/styled-components

💡 Three useState hooks could be combined into useReducer

✅ Proper PropTypes definitions

✅ Good accessibility attributes (aria-labels, roles)

⚠️  useEffect missing dependency 'userId' - will cause stale closures

Recommendation: Works, but needs refactoring before it grows larger.

Use Cases

1. Solo Developer - Be Your Own Code Reviewer

Working alone? No team to review your code? roast fills that gap.

$ roast src/new-feature.js

Benefit: Catch bugs, security issues, and bad patterns before they hit production. Get a second pair of (AI) eyes on your work.

When to use:

Before committing new features
After refactoring
When learning a new language/framework
Before deploying to production

2. Learning & Skill Building

Picked up a new language? roast teaches by critiquing.

# First Python script
$ roast my_first_app.py

# First Rust program
$ roast --serious main.rs

Benefit: Learn best practices, idioms, and common pitfalls by seeing them in your own code. Use --severity mild for encouragement while learning.

3. Pre-Commit Quality Gate

Add roast to your git hooks to enforce quality before commits.

# .git/hooks/pre-commit
#!/bin/bash
git diff --cached --name-only --diff-filter=ACM | \
  grep -E '\.(js|ts|py|go)$' | \
  xargs -I {} roast --serious {} || exit 1

Benefit: Automate code quality checks. Catch issues before CI, before code review, before your team sees them.

4. Onboarding to New Codebases

Just joined a team with a large codebase? Use roast to understand patterns.

$ roast src/core/auth-handler.js --serious

The AI will explain what the code does, highlight concerns, and suggest improvements. It's like having a senior dev walk you through the code.

Benefit: Faster onboarding. Understand existing patterns and anti-patterns.

5. Legacy Code Assessment

Inherited a legacy project? Get an honest assessment of what you're dealing with.

$ roast legacy/modules/payment-v1.php --severity harsh

Benefit: Prioritize refactoring. Know what's salvageable and what needs rewriting.

6. Teaching & Mentoring

Teaching someone to code? Show them what good vs. bad looks like.

# Student's code
$ roast student-assignment.py --severity mild

# Show them both roast and serious mode
$ roast --serious student-assignment.py

Benefit: Consistent, patient feedback. Teaches best practices without judgment.

7. Code Review Prep (Before Human Review)

Before requesting a PR review from teammates, run roast to catch obvious issues.

$ git diff main...feature-branch | roast --serious

Fix the issues roast finds, THEN request human review.

Benefit: Don't waste teammates' time on trivial issues. Submit cleaner PRs. Build a reputation for quality.

Performance Tips

1. Use Shell Aliases for Common Workflows

Create shortcuts for frequent roast scenarios:

# Add to ~/.bashrc or ~/.zshrc
alias roast-js='roast --serious'
alias roast-diff='git diff | roast --serious'
alias roast-mild='roast --severity mild'
alias roast-harsh='roast --severity harsh'

Now just type:

$ roast-diff  # Review uncommitted changes
$ roast-js src/app.js  # Serious review

2. Batch Review Multiple Files

Use shell loops for bulk reviews:

# Review all JavaScript files in src/
for file in src/**/*.js; do
  echo "=== $file ==="
  roast --serious "$file"
done

# Or with find
find src -name "*.py" -exec roast --serious {} \;

Save results:

for file in src/**/*.js; do
  roast --serious "$file" > "reviews/$(basename $file).review.txt"
done

3. Optimize API Costs

Roasting uses Anthropic's API. To reduce costs:

a) Review only changed files:

git diff --name-only | grep -E '\.(js|ts|py)$' | xargs -I {} roast {}

b) Use haiku model for simple reviews (modify source code):

// In lib/roast.js, change:
model: 'claude-haiku-4-5'  // Faster, cheaper, less detailed

c) Review smaller code sections:

# Instead of entire 1000-line file
$ head -100 large-file.js | roast

4. Cache Reviews Locally (Manual)

For files you review repeatedly:

# Save review for later reference
$ roast src/utils.js > reviews/utils-2026-02-08.txt

# Compare with previous review
$ diff reviews/utils-2026-02-01.txt reviews/utils-2026-02-08.txt

5. Combine with Git Hooks for Automation

Pre-commit hook (review before committing):

#!/bin/bash
# .git/hooks/pre-commit
git diff --cached --name-only | \
  grep -E '\.(js|ts)$' | \
  xargs -I {} roast --serious {}

read -p "Proceed with commit? (y/n) " -n 1 -r
echo
if [[ ! $REPLY =~ ^[Yy]$ ]]; then
  exit 1
fi

Pre-push hook (final check before pushing):

#!/bin/bash
# .git/hooks/pre-push
git diff origin/main...HEAD --name-only | \
  grep -E '\.(py|js|go)$' | \
  xargs -I {} roast --serious {} > /tmp/pre-push-review.txt

cat /tmp/pre-push-review.txt

6. Parallel Processing for Large Projects

Use GNU parallel for speed:

# Install: brew install parallel (macOS) or apt-get install parallel (Linux)

# Review all files in parallel (4 at a time)
find src -name "*.js" | parallel -j 4 roast --serious {} > reviews.txt

7. Smart File Selection

Don't roast everything. Focus on:

New/changed files:

git diff --name-only main | xargs -I {} roast {}

Complex files (high line count):

find src -name "*.js" -exec wc -l {} \; | \
  sort -rn | head -10 | \
  awk '{print $2}' | xargs -I {} roast {}

Files with recent bugs (from git history):

git log --format=format: --name-only | \
  sort | uniq -c | sort -rn | \
  head -10 | awk '{print $2}' | \
  xargs -I {} roast --serious {}

Troubleshooting

1. "ANTHROPIC_API_KEY not set" error

Problem:

$ roast src/app.js
💥 Error: ANTHROPIC_API_KEY environment variable is required

Solutions:

a) Set for current session:

export ANTHROPIC_API_KEY="sk-ant-your-key-here"

b) Make it permanent (add to shell config):

# For bash
echo 'export ANTHROPIC_API_KEY="sk-ant-your-key"' >> ~/.bashrc
source ~/.bashrc

# For zsh
echo 'export ANTHROPIC_API_KEY="sk-ant-your-key"' >> ~/.zshrc
source ~/.zshrc

c) Get your API key: Visit console.anthropic.com/settings/keys

2. "Cannot find file" error

Problem:

$ roast src/app.js
💥 Error: ENOENT: no such file or directory

Solutions:

a) Check the file path:

$ ls -la src/app.js  # Verify file exists

b) Use absolute path:

$ roast /full/path/to/src/app.js

c) Check current directory:

$ pwd
$ roast ./src/app.js

3. Empty or generic roast output

Problem:

$ roast my-code.js

🔥 CODE ROAST 🔥
──────────────────────────────────────────────────
This code looks fine.
──────────────────────────────────────────────────

Solutions:

a) File might be too small/simple:

# The code might genuinely be fine!
$ wc -l my-code.js  # Check line count

b) Language not detected. Rename file with proper extension:

$ mv script script.py  # Add .py extension
$ roast script.py

c) Use verbose mode (if available) or try serious mode:

$ roast --serious my-code.js

4. "Invalid severity level" error

Problem:

$ roast --severity extreme src/app.js
💥 Error: Invalid severity level: extreme
Valid options: mild, medium, harsh

Solutions:

a) Use valid severity:

$ roast --severity harsh src/app.js

b) Check available options:

$ roast --help

5. Command not found: roast

Problem:

$ npm install -g @muin/roast
$ roast src/app.js
zsh: command not found: roast

Solutions:

a) Check npm global bin directory is in PATH:

$ npm config get prefix
# Should show something like /usr/local or ~/.npm-global

$ echo $PATH
# Should include the bin directory from above

b) Fix PATH (add to ~/.bashrc or ~/.zshrc):

export PATH="$PATH:$(npm config get prefix)/bin"
source ~/.bashrc  # or ~/.zshrc

c) Use npx instead (doesn't require global install):

$ npx @muin/roast src/app.js

6. API rate limit errors

Problem:

$ roast src/app.js
💥 Error: rate_limit_error: You have exceeded your API rate limit

Solutions:

a) Wait a minute and retry (Anthropic has per-minute limits)

b) Check your rate limits: Visit console.anthropic.com/settings/limits

c) For batch processing, add delays:

for file in src/*.js; do
  roast "$file"
  sleep 3  # Wait 3 seconds between calls
done

d) Upgrade to higher tier if you hit limits regularly

7. Colored output broken / weird characters

Problem:

$ roast src/app.js
[31m🔥[0m [1mCODE ROAST[0m...

Solutions:

a) Disable colors:

$ roast --no-color src/app.js

b) Check terminal color support:

$ echo $TERM
# Should be xterm-256color or similar

# If it's 'dumb', set it:
export TERM=xterm-256color

c) Redirect to file (colors auto-disabled):

$ roast src/app.js > review.txt

8. Very slow responses (>30 seconds)

Problem:

$ roast very-large-file.js
# ...waiting forever...

Solutions:

a) File might be too large. Claude has token limits (~100k tokens ≈ 75k words).

Check file size:

$ wc -l very-large-file.js
5000 very-large-file.js  # Too large!

b) Split into smaller files:

$ split -l 500 large-file.js part-
$ roast part-aa
$ roast part-ab

c) Review only a section:

$ head -300 large-file.js | roast  # First 300 lines

d) Check Anthropic API status:

$ curl -I https://api.anthropic.com

9. Model not found error

Problem:

$ roast --model claude-opus-99 src/app.js
💥 Error: model 'claude-opus-99' not found

Solutions:

a) Use a valid model name:

$ roast --model claude-sonnet-4-5 src/app.js
$ roast --model claude-opus-4 src/app.js
$ roast --model claude-haiku-4-5 src/app.js

b) Check available models at docs.anthropic.com

c) Omit the flag to use default:

$ roast src/app.js  # Uses default model

10. Network/connection errors

Problem:

$ roast src/app.js
💥 Error: connect ETIMEDOUT

Solutions:

a) Check internet connection:

$ ping api.anthropic.com

b) Check if firewall/VPN is blocking:

$ curl https://api.anthropic.com

c) Set proxy if behind corporate firewall:

export HTTP_PROXY=http://proxy.company.com:8080
export HTTPS_PROXY=http://proxy.company.com:8080

11. Permission errors on Windows

Problem:

C:\> npm install -g @muin/roast
Error: EACCES: permission denied

Solutions:

a) Run as Administrator (PowerShell):

# Right-click PowerShell → "Run as Administrator"
PS> npm install -g @muin/roast

b) Or use npx (no install needed):

PS> npx @muin/roast src/app.js

12. File encoding issues (non-UTF8)

Problem:

$ roast legacy-code.php
💥 Error: Invalid UTF-8 sequence

Solutions:

a) Convert file to UTF-8:

$ iconv -f ISO-8859-1 -t UTF-8 legacy-code.php > legacy-code-utf8.php
$ roast legacy-code-utf8.php

b) Check current encoding:

$ file legacy-code.php
legacy-code.php: PHP script, ISO-8859 text

Frequently Asked Questions (FAQ)

1. Is my code sent to Anthropic's servers?

Yes. When you run roast, the file contents are sent to Anthropic's Claude API for analysis.

Security note: Do NOT roast files containing:

API keys, passwords, or secrets
Proprietary algorithms (if you have strict confidentiality agreements)
PII (personally identifiable information) in test data
Anything you wouldn't share with a third party

Anthropic's privacy policy: https://www.anthropic.com/legal/privacy

2. How much does it cost to use?

roast itself is free and open-source. However, it uses Anthropic's Claude API which has usage costs:

Free tier: $5 credit (enough for ~100-300 roasts depending on file size)
After that: ~$0.01-0.05 per roast (1-5 cents)
Monthly estimate: If you roast 20 files/day: ~$10-30/month

Check pricing: https://www.anthropic.com/pricing

3. Can I use this offline?

No. roast requires internet connection to reach Anthropic's API.

Alternatives for offline:

Set up a local LLM (complex, beyond scope)
Use traditional linters (eslint, pylint, etc.)
Manual code review

4. Is the humor offensive?

Roast mode: Sarcastic and blunt, but never personal. It criticizes code, not people.

Serious mode: Professional, no humor.

Severity levels:

Mild: Friendly and encouraging
Medium: Sarcastic but constructive (default)
Harsh: Very direct, can be brutal

Use --serious for team PRs or --severity mild for beginners.

5. Does it actually improve code quality?

Yes, when used correctly:

✅ Catches security issues (SQL injection, XSS, unsafe eval)
✅ Identifies performance problems (O(n²) loops, memory leaks)
✅ Highlights maintainability issues (long functions, tight coupling)
✅ Suggests modern best practices
✅ Teaches by explaining WHY something is bad

But:

AI can miss context-specific issues
It's a tool, not a replacement for human review
Always verify suggestions before applying

6. Can I customize the roast style?

Currently: Severity levels (mild/medium/harsh) and serious mode.

Not yet: Custom roast personas (e.g., "roast like Linus Torvalds" or "review like Uncle Bob").

Future: Customizable prompts/styles are on the roadmap.

7. What languages are supported?

✅ JavaScript / TypeScript / Node.js
✅ Python
✅ Go
✅ Rust
✅ Java / Kotlin
✅ C / C++
✅ Ruby
✅ PHP
✅ Swift
✅ Shell / Bash
✅ SQL
✅ HTML / CSS
✅ And more...

Claude has broad training, so even niche languages often get good results.

8. Does it work with frameworks (React, Django, etc.)?

Yes! The AI understands framework-specific patterns:

React: Hooks, component structure, props, state management
Django: ORM patterns, views, middleware
Express: Route handlers, middleware, error handling
Rails: ActiveRecord, controllers, migrations
etc.

It will roast framework-specific anti-patterns (e.g., prop drilling in React, N+1 queries in ORMs).

9. Can I use this for production code reviews?

Yes, but:

✅ Great for initial screening
✅ Catches obvious bugs and security issues
✅ Ensures code meets basic quality standards

⚠️ Should NOT replace human code review for:

Business logic validation
Architecture decisions
Team-specific conventions
Nuanced judgment calls

Best practice: Use roast --serious BEFORE requesting human review. Fix what it finds, then get human eyes on it.

10. Will it give different results each time?

Yes, slightly. AI models have some randomness (temperature setting).

Typically:

Core issues will always be flagged
Phrasing may vary
Severity of roast may differ slightly

For consistent results, use --serious mode which is more deterministic.

11. Can I contribute to the project?

Absolutely! Contributions are very welcome:

Report bugs via GitHub Issues
Suggest features
Submit pull requests
Improve roast quality
Add language support
Write tests

See Contributing section.

12. Does it store my code or reviews?

No local storage. roast doesn't save files or reviews on your machine (unless you manually redirect output).

API-side: Anthropic may temporarily log requests for abuse prevention. Check their privacy policy for retention details.

Tips

Share your roasts - They're designed to be screenshot-friendly
Use serious mode for PRs - Save the humor for your own code
Review before committing - Catch bugs before your CI does
Roast legacy code - Therapeutic and educational
Pipe from git diff - Review only what changed: git diff | roast
Works with stdin - cat sketch.py | roast or pbpaste | roast

Create project-specific wrappers:

# scripts/review.sh
#!/bin/bash
find src -name "*.js" -mtime -1 | xargs -I {} roast --serious {}

Combine with other tools:

# Lint first, then roast
$ eslint src/app.js && roast src/app.js

# Roast, then test
$ roast src/app.js && npm test

Team usage:

# Add to CI pipeline
- name: AI Code Review
  run: |
    git diff origin/main...HEAD --name-only | \
      grep -E '\.(js|ts|py)$' | \
      xargs -I {} npx @muin/roast --serious {}

Roadmap

v0.2 (Next Release)

JSON output mode - Structured output for CI integration
Config file support - .roastrc for defaults (severity, model, etc.)
Multi-file batch mode - roast src/*.js in one API call
Ignore patterns - .roastignore like .gitignore
Diff mode - Compare changes: roast --diff main...feature

v0.3

Custom roast personalities - "Roast like Linus", "Review like Uncle Bob"
Fix suggestions with diffs - Show exact code changes to apply
Performance scoring - Rate code quality 0-100
Security scan mode - Focus on vulnerabilities only
HTML report generation - roast --html src/ > report.html

v1.0 (Future)

Auto-fix mode - Apply suggested changes automatically
IDE plugins - VS Code, IntelliJ, Vim
Team dashboard - Track code quality over time
Learning mode - Personalized feedback based on your skill level
Multi-model support - Use GPT-4, Gemini, local models
Offline mode - Local model integration

Community Requests

Web UI for drag-and-drop file roasting
Slack/Discord bot integration
GitHub PR comment automation
GitLab CI/CD integration
Pre-commit hook generator
Support for Jupyter notebooks (.ipynb)

Want a feature? Open a GitHub issue or +1 existing ones!

Development

# Clone repo
git clone https://github.com/muinmomin/roast.git
cd roast

# Install dependencies
npm install

# Link locally for testing
npm link

# Test with sample file
roast examples/bad-code.js

# Run tests (if available)
npm test

Project structure:

roast/
├── bin/
│   └── roast.js         # CLI entry point
├── lib/
│   └── roast.js         # Core logic
├── examples/
│   └── bad-code.js      # Test cases
├── package.json
└── README.md

Testing:

# Create test file
cat > test.js << 'EOF'
var x = 1;
eval(userInput);
EOF

$ roast test.js

Contributing

Found a bug? Want to add a feature?

# Clone the repo
git clone https://github.com/muinmomin/roast.git
cd roast

# Install dependencies
npm install

# Test locally
npm link
roast test-file.js

# Run tests (if available)
npm test

Pull requests welcome! Please:

Keep the humor sharp but not mean
Add tests for new features
Follow existing code style (ESLint config included)
Update README if adding options

Code style:

Use ESLint
Follow existing patterns
Add JSDoc comments for functions
Keep roasts clever, not cruel

Ideas for contributions:

Better language detection
Framework-specific roasting (React, Django, Rails patterns)
Security vulnerability detection
Performance analysis
Custom roast templates

🛠️ More CLI Tools from MUIN

Love roast? Check out our other developer productivity tools:

Tool	Description	NPM
git-why	AI-powered git history explainer. Understand why code exists before you roast it.
portguard	Monitor and kill zombie processes hogging your ports. Fix `EADDRINUSE` in one command.
oops	Pipe any error to AI for instant fixes. When roasting reveals real bugs, `oops` fixes them.

Coming soon: More tools to make your dev workflow faster. Follow @muincompany for updates!

Featured On

📰 Read the launch article on Dev.to: 4 CLI Tools Every Developer Needs (That You've Never Heard Of)

License

MIT

Credits

Built with:

Anthropic Claude - The AI doing the roasting
Commander.js - CLI framework
Chalk - Terminal colors

Inspired by every brutally honest code review you've ever received (or given).

Built by MUIN 🚀

We build AI-powered tools that make developers' lives easier.

Website • Twitter • GitHub

Roast responsibly. Don't roast production code in public without permission.

Name		Name	Last commit message	Last commit date
Latest commit History 17 Commits
bin		bin
docs		docs
examples		examples
lib		lib
.gitignore		.gitignore
BUILD-SUMMARY.md		BUILD-SUMMARY.md
CHECKLIST.md		CHECKLIST.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
SETUP.md		SETUP.md
package-lock.json		package-lock.json
package.json		package.json
test-code.js		test-code.js

Folders and files

Latest commit

History

Repository files navigation

roast

🚀 Launching on Product Hunt — April 9, 2026

What is this?

Why use this?

How is roast-cli Different?

Installation

Setup

Usage

Roast mode (default)

Serious mode

Custom model

Supported Languages

Options

Severity Levels

Examples

Example 1: Quick file roast

Example 2: Serious mode for team PR review

Example 3: Review from stdin (pipe or paste)

Example 4: Git diff review before commit

Example 5: Custom model for complex code

Example 6: Mild severity for learning

Example 7: Harsh mode for reality check

Example 8: TypeScript React component

Example 9: Shell script review

Example 10: SQL query review

Example 11: Rust code review

Real-World Scenarios

Pre-commit sanity check

Reviewing that "quick fix" from 2am

Checking tutorial code before copy-paste

The "I learned this yesterday" review

Legacy code archaeology

Team code style check

Use Cases

1. Solo Developer - Be Your Own Code Reviewer

2. Learning & Skill Building

3. Pre-Commit Quality Gate

4. Onboarding to New Codebases

5. Legacy Code Assessment

6. Teaching & Mentoring

7. Code Review Prep (Before Human Review)

Performance Tips

1. Use Shell Aliases for Common Workflows

2. Batch Review Multiple Files

3. Optimize API Costs

4. Cache Reviews Locally (Manual)

5. Combine with Git Hooks for Automation

6. Parallel Processing for Large Projects

7. Smart File Selection

Troubleshooting

1. "ANTHROPIC_API_KEY not set" error

2. "Cannot find file" error

3. Empty or generic roast output

4. "Invalid severity level" error

5. Command not found: roast

6. API rate limit errors

7. Colored output broken / weird characters

8. Very slow responses (>30 seconds)

9. Model not found error

10. Network/connection errors

11. Permission errors on Windows

12. File encoding issues (non-UTF8)

Frequently Asked Questions (FAQ)

1. Is my code sent to Anthropic's servers?

2. How much does it cost to use?

3. Can I use this offline?

4. Is the humor offensive?

5. Does it actually improve code quality?

6. Can I customize the roast style?

7. What languages are supported?

8. Does it work with frameworks (React, Django, etc.)?

9. Can I use this for production code reviews?

10. Will it give different results each time?

11. Can I contribute to the project?

12. Does it store my code or reviews?

Tips

Packages