Question 1

If the report comes back clean, does that mean the project is safe?

Accepted Answer

No, and the README says so up front. It reads selected text files only: .env and .env.*, binaries, lockfiles, generated bundles and anything over 400,000 bytes are never uploaded, so malicious behavior in a skipped file simply does not appear. It does not sandbox code, and install scripts and postinstall hooks still run.

Question 2

Is this a replacement for a secret scanner or a dependency auditor?

Accepted Answer

Neither. It does not audit known dependency vulnerabilities and does not inventory committed secrets; the README points you at a dependency auditor and a secret scanner for those. What it adds is review of file contents for behavior across data theft, persistence, concealment, system abuse and build or CI supply-chain steps.

Question 3

Why do 0.91 and 0.09 get reported as equally confident?

Accepted Answer

That is a reporting artifact worth knowing. A noul answer carries no confidence field, so the report computes confidence as 2 * |p - 0.5|. Probabilities of 0.91 and 0.09 therefore get identical confidence while pointing to opposite answers, so read the probability rather than the confidence column.

Question 4

Will a PR from a fork get scanned, and could that leak my API key?

Accepted Answer

Fork PRs do not receive the API secret by default, and the README warns against switching to pull_request_target just to expose the key, since running untrusted PR code in that context can expose your secrets. Note that --diff-from scans whole changed files, not just changed lines.

Question 5

Which files does the scanner skip?

Accepted Answer

It reads selected text files and skips images, binaries, lockfiles, .env and .env.* files, dist/lib/build output, tsconfig and some JSON, and files above 400,000 bytes; it does not inspect binaries, installers, or running processes.

Question 6

How does the two-pass scan work?

Accepted Answer

The first pass sends each chunk's file contents with all check categories in one request, and a second pass runs on suspicious or uncertain chunks focusing on the line windows Jev identified, splitting and retrying if Jev returns max_tokens_exceeded.

Question 7

Can I point it at a different Jev-compatible service?

Accepted Answer

Yes. Set --base-url and --model, or TYPESAFE_BASE_URL, and the service must accept POST /v1/systemone with TypeSafe's state, model, and typed questions shape; token counts are reported but cost is left unknown for other providers.

Is Malicious

About this project

Highlights

Quickstart

Questions about this entry

Reactions & coverage

From the Jev guides

From the Jev AI model community