AI-assisted code security review

Find potential security issues. Review the evidence.

Give GRACE a scoped repository task. Follow its sandbox commands, inspect file-referenced findings, and review proposed fixes before using them.

AI-assisted review—not a security certification. Findings and proposed changes require human verification.

A review you can inspect

Follow the work, not just the conclusion.

A useful finding should explain where the issue is, why it matters, and what still needs verification.

  1. Scope the task

    Identify the repository area, relevant behavior, and constraints. Avoid an undefined “check everything.”

  2. Inspect the execution

    Follow commands and output as GRACE works in its sandbox. Check what was actually examined.

  3. Verify the result

    Review the findings and proposed changes. Reproduce important claims and run appropriate checks yourself.

A better first task

Use a brief that asks for evidence.

Copy this request into your workspace after supplying the code and context you are authorized to share.

Review the supplied repository for potential security issues.

Scope:
- Focus on authentication, authorization, and session handling.
- Explain assumptions about deployment and trust boundaries.
- Do not change unrelated code.

For each potential finding:
- Reference the relevant file and code.
- Explain the risky behavior and conditions required.
- Distinguish confirmed evidence from uncertainty.
- Propose a focused fix and explain its trade-offs.

Deliver:
- A Markdown findings report.
- Proposed changes, where appropriate.
- Commands and checks performed, including failures.
- Checks not run and remaining limitations.

Do not claim the repository is secure because no issues were found.

Include the context a reviewer would need

Explain the framework, authentication model, deployment assumptions, and commands for running tests. Do not include production credentials or customer data.

Before you start

Useful assistance. Clear limits.

Treat this as another review input, not permission to deploy unchecked changes.

  • Potential findings, not guaranteed vulnerabilities

    Confirm reachability, exploitability, and real deployment conditions before assigning severity.

  • A clean report does not prove the code is secure

    Issues can be missed. Use appropriate tests, specialist tools, and independent review.

  • Review only code you are authorized to provide

    Remove secrets and sensitive data. Check your organization’s requirements before sharing code.

Common questions

Know what you are starting.

What does GRACE do during a code review?

GRACE can inspect repository files, run commands in its sandbox, and produce findings or proposed changes. Review the transcript and output to understand what happened during your particular task.

Does this replace a penetration test?

No. An AI-assisted code review is not a penetration test, compliance assessment, or security certification.

Will a proposed fix be safe to deploy?

Not automatically. Review the diff, reproduce the relevant behavior, run tests, and follow your normal approval process before deploying.

Where can I check access and pricing?

Check the pricing page for current plans and allowances. Sign in to see the access available to your account.

Start small

Bring one security question about your code.

Define the scope. Ask for evidence. Review the result.

Start a code review