credsweep

Quick Start

Install via jcli (recommended)

jcli install credsweep

If you don't have jcli yet, install it first with curl -fsSL https://cli.johlem.net/tools/jcli/install.sh | bash.

Scan a codebase

credsweep --path ./src                          # walk a directory
credsweep --stdin < suspicious.log              # read from stdin
credsweep --git --path .                        # scan git history for secrets ever committed
credsweep --path . --output sarif > cs.sarif   # SARIF for GitHub Code Scanning
credsweep --install-hook --path .               # write a .git/hooks/pre-commit

What it does

credsweep is a credential / secret scanner — feed it a file tree, stdin, or a git repository's history and it returns every accidentally committed AWS key, GitHub PAT, private key, JWT, database connection string, and ~45 other things. Built in Rust, single binary. Absorbed the retired credaudit tool — its full 53-pattern catalogue plus its sensitivity filter, entropy validation, git-history scan, and false-positive filtering now live here.

• 53 patterns across 9 kinds — cloud_credential, token, api_key, connection_string, private_key, webhook, password, secret, auth_header. AWS, GitHub, Slack, Stripe, Google, Azure, Mailgun, Datadog, Cloudflare, Vault, Terraform Cloud, Heroku, Twilio, SendGrid, DigitalOcean, Shopify, GitLab, NPM, Docker Hub, and the generic patterns. All embedded in the binary via include_str!.
• Three sensitivity levels. --sensitivity high runs only prefixed/high-confidence patterns (lowest FP rate). medium (default) adds DB connection strings, JWTs, bearer tokens. low adds the generic password = "..." heuristics (highest yield, highest FP rate).
• Value-scoped false-positive filtering. Generic-assignment patterns capture the value in group 1; entropy + FP-keyword checks run against the value, not the whole match. So password = "password" is suppressed but password = "<real value>" fires.
• Shannon-entropy floor per pattern. Each pattern declares an entropy_min; matches below the threshold are dropped as likely placeholder (xxxxxxxx, your_api_key_here, etc.).
• Git history scan. --git walks every commit's diff (with --root so initial commits are included) and stamps the SHA on every finding. Catches credentials that were committed and later removed.
• SARIF v2.1.0 output. --output sarif emits GitHub Code Scanning-compatible JSON with stable partialFingerprints so GitHub deduplicates findings across runs.
• Baseline / update. --baseline snapshots current findings; --update only reports new ones. Ship credsweep against a codebase with existing known leaks without flooding CI.
• Allow-list. --allow-list <FILE> reads a TOML file with file_glob / pattern / snippet matchers — all ANDed. Documented test fixtures get a clean pass without disabling patterns globally.
• Pre-commit hook installer. --install-hook writes .git/hooks/pre-commit (respecting core.hooksPath for monorepos). Refuses to clobber an existing hook unless --force.

CLI

Subcommand: credsweep patterns list

Show every embedded detection pattern with its severity, sensitivity, kind, and confidence. Useful for confirming what the scanner is looking for before committing to a sensitivity tier, or building per-team allow-lists from the catalogue.

credsweep patterns list                                # full catalogue (text)
credsweep patterns list --severity critical            # filter by severity
credsweep patterns list --sensitivity high --format md # markdown table
credsweep patterns list --format json | jq '.[].name'  # pipe to jq

Flags: --format {text,json,md}, --severity {info,low,medium,high,critical}, --sensitivity {high,medium,low}.

Allow-list TOML format

# All present matchers are ANDed.
# At least one of (file_glob, pattern, snippet) must be set.

[[allow]]
file_glob = "tests/**"
pattern   = "JWT Token"
note      = "test fixtures — checked manually"

[[allow]]
snippet = "AKIAJSIRBVE5YQQTESTX"
note    = "intentional fixture key in docs"

[[allow]]
pattern = "Generic Password Assignment"
note    = "this codebase's password=... lines are template variables"

CI workflow

Common pattern: baseline once, then update on every PR. Pre-existing leaks don't break CI; new ones do.

# One-time setup (committed alongside the codebase)
credsweep --path . --sensitivity medium --baseline
git add .credsweep-baseline.json
git commit -m "ci: credsweep baseline"

# CI step (.github/workflows/ci.yml etc.)
credsweep --path . --sensitivity medium --update --output sarif --out-file cs.sarif

Design choices

• Subprocess to git, not git2. The --git scanner shells out to git diff-tree -p --no-commit-id --root -r <commit>. The --root flag is critical — without it, the initial commit (no parent) produces empty diff output and the credential introduced there is missed. Saves ~10 MB of binary vs vendoring libgit2.
• Snippet redaction is mandatory. The full matched string never leaves the binary. "AKIA1234567890ABCDEF" becomes "AKIA…CDEF" in every output mode. Operators can grep their codebase for the same pattern but can't accidentally paste a real secret into logs / Slack / a JIRA ticket.
• Pipe-clean structured output. When --output json|csv|markdown|sarif is set without --out-file, the banner and per-finding terminal display are suppressed so stdout is pure structured output.
• Exit codes. 0 = clean, 1 = findings, 2 = runtime error, 3 = bad arguments. Suitable for CI gating.