The Dangerous Part Is Rarely the Source Code
It is a postinstall script, a leaked token, a Git hook, an unpinned action, a hidden instruction in a config file. Pentesterra analyses all of it before you push - dependencies, secrets, CI/CD, IaC, and AI toolchain - without your code ever leaving your machine.
What gets analysed
Supply chain and dependencies
Every dependency is checked against CVE and CISA KEV data, malicious-package intelligence, and typosquat patterns. Lifecycle scripts, postinstall hooks, and install-time behaviour are inspected - the part of a package that runs before anyone reads the code.
- →CVE / KEV mapping with exploitability context
- →Known-malicious and typosquatted package detection
- →postinstall / lifecycle-script and install-behaviour review
- →SBOM generation and SARIF export
Secrets and credential flow
Secrets are found by pattern and entropy, then fingerprinted with a SHA-256 hash - never the value itself. Credential flow analysis catches proxy redirects and environment-variable exposure, including the MCP-server case where an integration can read your keys.
- →Pattern + entropy secret detection, values never transmitted
- →Token liveness and blast-radius assessment
- →Proxy-redirect and env-exposure credential flow checks
- →Commit-message and shell-history scanning
Repository, CI/CD, and IaC
Malicious repositories rarely look malicious in the source. Pentesterra checks Git hooks, config, filters, and history, CI/CD workflow injection, and infrastructure-as-code misconfiguration before you run or push unknown code.
- →Git hook / config / filter / history inspection
- →CI/CD workflow injection and secret-exfil patterns
- →Terraform / IaC misconfiguration checks
- →SVN history exposure scanning
AI toolchain and logic
The developer surface now includes AI tooling. Pentesterra flags risky MCP configuration, unrestricted agent permissions, prompt-injection in data files, and torch.load-style RCE, plus SAST-lite logic checks and API-route security - mapped to the OWASP LLM Top 10.
- →MCP / IDE-extension / agent-config risk
- →Prompt injection in .md / .json / .csv / .yaml data files
- →torch.load RCE, .pth runtime-hook, RAG SSRF and tenant-isolation checks
- →SAST-lite logic flaws and API-route auth issues
FAQ
Does my source code leave my machine?
No. The local agent collects only structured metadata - dependency types, paths, secret fingerprints (SHA-256, never the value), and endpoint metadata - and sends a redacted payload to the cloud engine. Source code never leaves the machine.
Where does it run?
As a standalone IDE plugin for VS Code, Cursor and Windsurf that a developer runs before pushing, as a CLI, or wired into CI/CD for automated gate enforcement. No pipeline integration is required to start.
How is this different from a normal SAST or SCA tool?
Most tools look at one layer - source code, or dependency versions. Pentesterra also inspects install-time behaviour, repository configuration, CI/CD, IaC, secrets flow, and AI toolchain risk, and it correlates code-level risk with production exploitability from the rest of the platform.
Is there a free option?
Yes. DevGuard, which delivers this capability at the IDE and CLI level, is free to try with no CI/CD integration or account changes required.