keyburn 0.1.5

Secret scanner for the AI coding era — catches leaked API keys before they ship.

keyburn

Secret scanner built for the AI coding era.

Cursor, Copilot, and Lovable write code fast. Sometimes too fast — hardcoded API keys, database URLs, and JWT secrets end up committed before anyone notices. Keyburn catches them before they ship.

• 57 detection patterns — AWS, OpenAI, Anthropic, Stripe, Supabase, GitHub, Slack, and more
• Shannon entropy analysis — catches secrets that don't match a known pattern
• Actionable remediation hints — tells you exactly how to fix each finding, not just that something's wrong
• Framework-aware — knows that NEXT_PUBLIC_SECRET_KEY is exposed to the browser, that Supabase service_role keys bypass RLS, etc.
• Zero noise escape hatches — # keyburn:ignore, allowlists, baselines, per-path excludes
• CI-native — text/JSON/SARIF output, GitHub code scanning integration, fail threshold per severity

---

Install

pip install keyburn
# or, zero-install:
pipx run keyburn scan .

Quickstart

# Scan current directory
keyburn scan .

# Scan and fail CI on any high-severity finding
keyburn scan . --fail-on high

# Only scan files changed in this PR
keyburn scan . --diff origin/main

# Only scan staged files (pre-commit)
keyburn scan . --pre-commit

# Output SARIF for GitHub code scanning
keyburn scan . --format sarif --out keyburn.sarif

Verify likely-live keys

keyburn verify checks whether a detected token is still active for selected providers.

Supported providers (current): openai, github, stripe.

# safest path: load from env var (avoid shell history leaks)
keyburn verify --from-env OPENAI_API_KEY --provider openai

# auto-infer provider from token format
keyburn verify --from-env GITHUB_TOKEN --provider auto

# fail CI if token appears active
keyburn verify --from-env STRIPE_SECRET_KEY --provider stripe --fail-on-valid

GitHub Action

Add this to any workflow — it scans on every push/PR and uploads findings to GitHub code scanning:

- name: Scan for secrets
  uses: safouanb/keyburn@v0

Or with options:

- name: Scan for secrets
  uses: safouanb/keyburn@v0
  with:
    fail_on: high          # low | medium | high (default: high)
    format: sarif          # text | json | sarif (default: sarif)
    upload_sarif: "true"   # upload to GitHub code scanning (default: true)
    comment_pr: "true"     # sticky PR comment summary (default: false)

Full workflow example:

name: Secret scan

on: [push, pull_request]

jobs:
  secrets:
    runs-on: ubuntu-latest
    permissions:
      security-events: write  # required for SARIF upload
      pull-requests: write    # required if comment_pr=true
    steps:
      - uses: actions/checkout@v4
      - uses: safouanb/keyburn@v0
        with:
          fail_on: high
          comment_pr: "true"

Pre-commit hook

# .pre-commit-config.yaml
repos:
  - repo: https://github.com/safouanb/keyburn
    rev: v0.1.5
    hooks:
      - id: keyburn

Or use the CLI directly in a git hook:

# .git/hooks/pre-commit
#!/bin/sh
keyburn scan . --pre-commit --fail-on high

Suppressing false positives

Inline, for a single line:

EXAMPLE_KEY = "not_a_real_secret_for_docs"  # keyburn:ignore

Allowlist by regex in keyburn.toml:

[allowlist]
regex = [
  "^sk_test_",        # all Stripe test keys
  "^EXAMPLE_KEY_",   # placeholder values in docs
]

Baseline — accept current state, catch new ones:

# Run once to write the baseline
keyburn scan . --baseline baseline.json --update-baseline

# Future scans only alert on NEW findings
keyburn scan . --baseline baseline.json --fail-on high

Skip files or directories:

[scan]
exclude_paths = ["tests/fixtures/**", "docs/examples/**"]
exclude_dirs  = [".venv", "node_modules"]

Disable specific rules:

[scan]
disable_rules = ["stripe-secret-test", "entropy"]

Configuration (keyburn.toml)

Drop a keyburn.toml in your repo root (see keyburn.toml.example):

[scan]
max_file_size_bytes = 2097152
exclude_dirs        = [".venv", "node_modules", "dist"]
exclude_paths       = ["tests/fixtures/**"]
disable_rules       = []
respect_gitignore   = true   # skip files in .gitignore (default: true)

[allowlist]
regex = []

What it detects

Category	Providers
AI APIs	OpenAI, Anthropic, Groq, HuggingFace, Replicate, Cohere
Cloud	AWS (key + secret + session), GCP, Firebase
Payments	Stripe (live + test + restricted)
Auth	GitHub PATs (classic + fine-grained), OAuth tokens, Clerk, Auth0
Comms	Slack (tokens + webhooks), Twilio, SendGrid, Mailgun, Discord, Telegram
Database	PostgreSQL, MySQL, MongoDB, Redis (connection strings)
Infra	Heroku, Vercel, Netlify, Doppler, Sentry
E-commerce	Shopify
Packages	npm, PyPI tokens
Framework	NEXT_PUBLIC_ secrets, Supabase service role vs anon key
Generic	Hardcoded passwords, JWT secrets, PEM keys, .env pasted into source
Entropy	High-entropy strings assigned to secret-looking variables

Output formats

Text (default) — rich panels with remediation hints per finding:

╭─ HIGH  Stripe Secret Key (live) ────────────────────────────────╮
│   File: src/payments.js:12:20                                    │
│   Rule: stripe-secret-live                                       │
│   Match: sk_l****************************1234                    │
│                                                                  │
│   How to fix: This is a LIVE Stripe key — it can charge real     │
│   cards. Roll it immediately at dashboard.stripe.com/apikeys.    │
│   Use STRIPE_SECRET_KEY env var instead.                         │
╰──────────────────────────────────────────────────────────────────╯

JSON — machine-readable with full finding metadata:

keyburn scan . --format json | jq '.summary'

SARIF — GitHub code scanning compatible:

keyburn scan . --format sarif --out keyburn.sarif

Git history scanning

Scan the last N commits for secrets that were added then "deleted":

keyburn scan --history 50    # last 50 commits
keyburn scan --history all   # full history (slow on large repos)

Precision benchmark

Track scanner noise over time against real OSS repositories:

python scripts/benchmark_precision.py

Results are written to:

• benchmarks/results/latest.json
• benchmarks/results/latest.md

CI can run this weekly via .github/workflows/precision-benchmark.yml.

Adoption loop

Track external adoption/noise metrics across real open-source repositories:

python scripts/adoption_loop.py

Results are written to:

• adoption/results/latest.json
• adoption/results/latest.md

CI can run this weekly via .github/workflows/adoption-loop.yml.

Known limitations

• Key verification currently supports only OpenAI, GitHub, and Stripe.
• Verification is a live HTTP check and can return unknown during rate limits/outages.
• A valid check means "accepted by provider now"; it does not measure scope or blast radius.
• .gitignore matching is intentionally lightweight and may differ from full git pathspec semantics.

Open-core

The scanner and all detection rules are open-source (Apache-2.0). Dashboards, Slack/email alerting, and one-click credential rotation live in a separate hosted product.

Contributing

See CONTRIBUTING.md. New detectors: open an issue with the provider, a redacted example, expected severity, and known false positives.

License

Apache-2.0. See LICENSE.