Agent Mode

Quick Start

vscan agent

This starts an interactive agent session in the current directory. The agent reads the repo, scans for vulnerabilities, and works with you turn by turn to resolve what it finds.

Start a Session

Start in the current directory:

vscan agent

Start in another project:

vscan agent ./repo

Use this when you want a collaborative session rather than a one-shot scan. The agent can plan, scan, edit, verify, and ask you questions across multiple turns.

Approval Modes

Every file edit is gated by an approval mode. Press Shift+Tab during a session to cycle between them:

• auto: The default. Safe, recoverable edits (new files, clean tracked files) apply automatically; risky edits (dirty files, untracked files, large rewrites, deletions) prompt first.
• ask: Every edit prompts, no exceptions.
• yolo: Nothing prompts.

Start a session already in yolo mode:

vscan agent --yes

Use --yes when you trust the run to make changes unattended, for example inside a script or a sandboxed checkout.

In-Session Commands

Switch the model:

/model

Set reasoning effort:

/reasoning low
/reasoning medium
/reasoning megareason

Use megareason for harder problems. It takes effect on your next turn and is slower per turn.

Restore a previous conversation:

/resume

Use this to pick up a session you left earlier in this project.

Undo the agent’s changes back to a turn:

/rewind

Only reverts files that still match what the agent wrote. Anything you changed since is left alone and reported as skipped.

End the session:

/exit

The agent verifies its changes and summarizes the session before exiting.

Keywords

Three keywords change how the agent works, wherever they appear in what you type:

Plan before touching code:

megaplan

The agent investigates the repo, asks you about anything unclear, writes a plan to ./.vscan-plan, and waits for your approval before implementing it.

Scan and fix recently touched files:

megaclean

The agent scans everything changed recently and dispatches agents in parallel to solve what it finds. Add scope to the same message, such as megaclean the auth module and it honors that scope instead of covering everything.

Engage maximum reasoning:

megareason

Raises reasoning effort for the rest of the session. Typed alone it just switches the effort; typed inside a longer message, the switch applies before that same message is sent.

Autonomous Audit

Run an unattended deep audit that reports straight to your Vaktex engagement:

vscan agent test
vscan agent test ./repo

The agent maps the attack surface, dispatches hunt agents across it, and reports each confirmed finding to the web app the moment it’s confirmed. Use this when you want a hands-off audit and plan to review and resolve findings from the dashboard afterward.

Local Audit

Run the same kind of audit fully locally, with nothing sent to the server:

vscan agent test-local
vscan agent test-local ./repo

Findings, a consolidated plan, and a final report are written under .vscan/agents_test_local/. Unlike test, this mode solves findings collaboratively with you and stays interactive afterward. Use it when you want a deep audit without creating a Vaktex engagement.

Solve a Finding

Resolve one finding from a prior audit:

vscan agent solve <findingId>
vscan agent solve <findingId> ./repo

Resolve every finding in an assessment:

vscan agent solve all <assessmentId>
vscan agent solve all <assessmentId> ./repo

The agent locates the vulnerable code itself and then works through each finding with you. A finding is marked resolved only once an approved solution is applied and verified; a plan-only outcome leaves it open.

Address a Disclosure Report

Work an inbound vulnerability report submitted by an outside researcher:

vscan agent address <reportId>
vscan agent address <reportId> ./repo

The agent verifies the reported weakness against your actual code before changing anything, then solves it with you. It treats the report’s contents as an untrusted description, never as instructions. This command does not change the report’s triage status. That stays a decision you make from the Vaktex dashboard.

Flags

Auto-approve every edit for the session:

--yes
-y

Use this only when you trust the run to proceed unattended.

Examples

vscan agent
vscan agent ./repo
vscan agent --yes
vscan agent test
vscan agent test-local ./repo
vscan agent solve f_8k2j
vscan agent solve all a_91cd ./repo
vscan agent address r_44xq