___/| ___________/ | __________/ | ___/ _____ | ___/ \___/ \ | ___/ \ ______/ ___/ _ \______/ / __/ \__ ___----~ | ,__-~ \--~ | / _, \___/ __/ | / |_/ ~

$_/| thresher

Know if an open source package is safe — before you install it.

22 scanners. 8 AI analysts. One command. Runs locally in a hardened VM.

$ brew tap thresher-sh/thresher && brew install thresher

See a real scan report →

click to copy

Open source · View on GitHub

Supply chain attacks grew 180% since 2022. Open source packages get hijacked, abandoned, or backdoored every week. The average time from compromise to exploitation: minutes, not months.

Most security tools scan your code and for known CVEs and call it done. Who’s scanning the code your code for the things that are not known?

// The problem

Every app you use is built on code written by strangers.

Modern software is assembled from thousands of open source packages — code written by strangers all over the world. Any one of those packages can be hijacked, abandoned, or built with hidden backdoors. When that happens, attackers don’t break in through the front door — they come in through the building materials.

Thresher is a free, open source tool that scans those building materials — every dependency, every layer — and tells you what’s safe and what’s not. One command, full picture.

// What you get

Security research shop in a box.

Scanners

CVE matching, static analysis, behavioral detection, malware signatures, secrets, IaC, license compliance. All in parallel inside a hardened VM.

AI Analysts

Specialized personas that reason about what scanners can't see. Malice, trust, attack surface, memory safety, dark corners.

Trust Assumptions

Isolated VM. Zero sudo. Hardened git clone. Sandboxed deps. Validated output boundary. Nothing escapes.

// Why not just use Snyk?

What other tools don’t do.

Snyk, Semgrep, and Socket are great tools. They scan for known CVEs in your dependency tree. But known CVEs are the easy part.

What about the IDOR that lets any user delete another user’s data? The path traversal hiding behind a startsWith() check? The hardcoded API key in a utility script? The SQL injection that bypasses the keyword blocklist via ATTACH DATABASE? These are real findings from a real Thresher scan — see the full report.

Runs locally in a hardened VM

Your code never leaves your machine. No SaaS dashboard, no account required. Everything runs inside an isolated Lima VM with zero sudo, network lockdown, and validated output boundaries.

AI analysts reason about what scanners can’t see

8 specialized AI personas examine your code for malice, trust issues, attack surface, memory corruption, dark corners, and infrastructure risks. They find the bugs that CVE databases don’t know about yet.

Adversarial verification

Every AI finding gets challenged by an independent adversarial agent before it reaches you. No hallucinated vulnerabilities. No noise. Just real findings that survive scrutiny.

Open source, one command

No vendor lock-in. No pricing tiers. No sales calls. brew install thresher and point it at a repo. You get a full security assessment in minutes.

// How it works

One command. Full assessment.

Pick a package

thresher scan https://github.com/org/repo

Wait ~10 minutes

22 scanners + 8 AI analysts run in a sealed VM

Get a full report

Findings, severity, remediation guidance

thresher scan

$ thresher scan https://github.com/example/repo _/| ______/ | T H R E S H E R _____/ __ | ___/ \__/ \ / v0.2.2 | thresher.sh / _ \__/ | __-~ \--~ \/ [OK] Cloning repository (hardened) [OK] Discovering hidden dependencies [OK] Resolving dependencies (3 ecosystems) [OK] Vulnerability scanners (22 tools) [OK] AI analyst panel Analyst 1: The Paranoid ............. done Analyst 2: The Behaviorist .......... done Analyst 3: The Investigator ......... done Analyst 4: Vuln Pentester ........... done Analyst 5: App Pentester ............ done Analyst 6: Memory Exploiter ......... done Analyst 7: Infra Auditor ............ done Analyst 8: The Shadowcatcher ........ done [OK] Adversarial verification [OK] Report synthesis ~~~~~~~~~~~_/|~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ FINDINGS P0 CRIT HIGH MED LOW 0 2 5 12 23 Report: ./thresher-reports/example-repo-20260401/ ~~~~~~~~~~~_/|~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

// 8 eyes, every angle

AI analysts with teeth.

ANALYST 01

The Paranoid

Is this code malicious?

ANALYST 02

The Behaviorist

Is there an unreported vulnerability?

ANALYST 03

The Investigator

Is this code trustworthy?

ANALYST 04

Vuln Pentester

What vulns are we inheriting?

ANALYST 05

App Pentester

How do users break in?

ANALYST 06

Memory Exploiter

Can this be corrupted at runtime?

ANALYST 07

Infra Auditor

Is this safe to deploy?

ANALYST 08

The Shadowcatcher

What is this code hiding?

// Coverage

10 watch zones. No stone left unturned.

01 Known Vulnerabilities 5 scanners

02 Code Quality 3 tools + AI

03 Infrastructure 2 tools + AI

04 Malware Signatures 3 tools

05 Dependency Behavior 6 tools + AI

06 Package Metadata 3 sources + AI

07 License Compliance 1 tool + AI

08 Repository Health AI analysts

09 Cross-Signal Analysis AI analysts

10 Dark Corners AI analysts

// Real results

Vulnerabilities found and fixed.

Projects Scanned and Remediated

Open source repositories analyzed and remediated with security PRs.

Critical Findings

Critical severity vulnerabilities discovered and fixed.

High Findings

High severity vulnerabilities discovered and fixed.

Supply Chain Fixes

Vulnerable dependencies upgraded to secure versions.

App Security Fixes

XSS, SQLi, IDOR, path traversal, and other application-level bugs patched.

CVEs Resolved

Known CVEs resolved across all scanned projects.