Audit

The audit and the fix are the same session.

Scanning where the code is, rather than reporting to somewhere the code is not.

Download for macOS
v0.1.33 · .dmg · Apple Silicon & Intel
Signed & notarized by Apple · opens without a Gatekeeper warning
sha256 698955a0187bc039f4c74f5d05a9f10fbb27376a45788a0a241d1326b73873c7
Download for Windows instead
$curl -fsSL https://lucidtrain.com/install.sh | sh

Most dependency scanning lives somewhere other than where the fix gets written: a CI job, a dashboard, a bot pull request. Lucid Train runs the audit in a tab of the application that already has the repository open, on a persona that starts in plan mode with read and shell tools but no ability to write. Clicking Apply fixes is what grants editing tools and switches the same session into making the changes. The gap between reading a finding and having a reviewed diff is one click rather than a ticket.

Plan first is the default, not a setting

The audit persona starts in plan mode, which has read access, shell access, web search and the audit tool but no file writing at all. It can therefore investigate a finding, check what depends on the package and propose an upgrade path without being able to touch anything. Editing tools appear only when you explicitly ask for the fixes.

Context the report does not have

A scanner knows a package has an advisory. It does not know that the package is only used in one script, or that the upgrade crosses a major version that changes an interface you rely on. An agent with the repository open can go and look, which is the difference between a list of findings and a decision about which ones matter here.

Reviewable, and refusable

Every edit arrives as a diff and every command is risk-tiered before it runs: safe ones run, riskier ones prompt, and the catastrophic tier is refused outright. Automating the tedious half of dependency upgrades is only worth doing if the review step is real, so the same approval machinery the coding tab uses applies here unchanged.

Available to other engines too

The audit is exposed as a tool over Lucid Train's own MCP server, so it arrives inside a Claude Code, Codex, Cursor CLI or OpenCode session as mcp__lucid__security_audit and is pre-approved rather than prompting. Running your agent of choice does not mean losing this.

FAQ

Questions

Can the audit change files on its own?

Not until you ask. It starts in plan mode with no writing tools, and Apply fixes is what grants them for that session.

How is this different from a bot that opens upgrade pull requests?

A bot proposes the bump without knowing your codebase. This has the repository open and the same session can check what the change would break before proposing it. It also produces a diff in your checkout rather than a PR to triage.

Does it work if I use a different coding agent?

Yes. The audit is available over MCP as mcp__lucid__security_audit inside Claude Code, Codex, Cursor CLI and OpenCode sessions.

Related

Last updated