Found no new or pre-existing flaws. You will only see LeoTrace again when it finds something.
Added GET /invoices/export. It writes the customer’s invoices to a CSV file in the exports folder.
Reviewed 1 changed file and flagged 1 finding to fix below.
Path traversal in billing/export.py:14: the file name comes straight from the request, so ../ can walk out of the exports folder. Keeping the path inside it now.
Fixed. The path stays inside the exports folder now.
Your agent’s security advisor.
There for every change.
New flaws stopped, existing ones fixed at source and your scanner backlog cleared, by your agents as they work.
Claude Code
Codex
GitHub Copilot
Cursor
Devin
Gemini CLI
- Antigravity
Windsurf
OpenCode
- Hermes Agent
One prompt to set up. Then every change is secure.
LeoTrace runs inside the coding agent you already use, so nothing in your pipeline changes.
- 01
Your agent finishes its turn. LeoTrace reads exactly what it changed, and the code that change reaches.
- 02
It picks the rules that apply to that change, written from real incidents.
- 03
A security judge reads the whole file and decides.
- 04
Anything it finds goes back to your agent with the reason, before you see done.
- 05
When your agent has acted, LeoTrace reads the change again.
Five jobs, one advisor.
Three as your agent works, two at scale.
Your agent writes code. No new flaw goes with it. LeoTrace checks the code your agent just wrote. If there is a flaw, the agent fixes it before you see done.
Added GET /invoices/export. It writes the customer’s invoices to a CSV file in the exports folder.
Reviewed 1 changed file and flagged 1 finding to fix below.
Path traversal in billing/export.py:14: the file name comes straight from the request, so ../ can walk out of the exports folder. Keeping the path inside it now.
Fixed. The export now builds the path with safe_join, so it stays inside the exports folder, and the endpoint works as before.
The path traversal at billing/export.py:14 is fixed.
The flaws only a security expert would catch. LeoTrace spots the flaws already in the files your agent touches, and advises your agent to fix them in the same turn when it is safe.
Added page and page_size to GET /orders, with 50 orders a page by default.
Reviewed 1 changed file and raised 2 findings. See below.
SQL injection in orders/list.py:31: the sort order comes straight from the query string. It was already there, and it is safe to fix inside this change, so fixing it now.
Done. The sort column now comes from an allowed list, and an order is only returned to the account that placed it.
Both findings in orders/list.py are fixed.
Your scanner findings, cleared as you work. Connect the AppSec tool you already use. Real findings are fixed in the turn, and the rest are closed as false positives, with the reason.
Added a search box above the customer list. It filters by name and email as you type.
Reviewed 2 changed files and raised 1 finding, plus 3 findings from your AppSec tool to double-check. See below.
I agree all 3 disputed findings are false positives.
customers/query.ts:42, where the search term is joined into the query. It was already there, and it is a small safe fix inside this change, so fixing it now.The SQL injection at customers/query.ts:42 is fixed. The 3 false positives are ignored in your AppSec tool, each with the reason.
Your whole backlog, in one sweep. Run one command and your agent works through every open finding, with LeoTrace confirming each one.
Running the sweep sends the full contents of every file with an open scanner finding to LeoTrace for judging. Secret files and symlinks are never sent.
Checked 25 open scanner findings in 14 files: 23 look like false positives, 2 are real flaws.
Now checking the 23 disputed findings.
I agree with all 23 suspected false positives.
docs/examples/ never run. The docs site only shows them as text.staticData.ts only ever loads fixed file names, never user input.test/ are left out of the build.2 real flaws, both small fixes:
.github/workflows/ci.yml:198: pin coverallsapp/github-action@v2 to a commit SHA (medium).github/workflows/ci.yml:351: add persist-credentials: false to the checkout step (low)Fixing the 2 small flaws now.
Rechecked the last sweep’s real findings: 2 fixed, 0 still open.
Both are fixed: coverallsapp/github-action is pinned to a commit SHA, and the deploy checkout sets persist-credentials: false.
Prove what agents wrote. Reviewed inside the agent, before a PR exists. Then the record: how much of your code AI wrote and which files we reviewed, by agent and engineer.


Ship knowing every change was checked.
A security advisor checks every change your agent makes, so a flaw is fixed while you are still in the file, not in a ticket next sprint.

Your judgement, in every agent.
LeoTrace applies AppSec judgement at the moment code is written, so your team spends its time on the findings that need a person.


Every change attested.
Down to the file.
Reviewed in the turn, not at the PR.


Roll it out across every team. With one standard for every agent.
Match your org tree, sign in with SSO and SCIM, set policies per workspace and keep a full audit trail.
Org tree
Workspaces that match how you report.
SSO and SCIM
People added and removed by your identity provider.
Policies
Rules for each workspace, set once.
ISO 27001
Certified, with a Trust Centre for your review.
Questions engineers ask
Only on turns that changed code, and only while the review runs.
Any model your agent runs. LeoTrace reviews the code, not the model.
Tell the agent. An answered finding is not raised again for 30 days.
Yes. On a turn that changed code, the changed files, the local helpers they import and your prompt go to LeoTrace for review. A sweep, only when you run it, sends the whole files that hold open scanner findings. The Trust Centre lists exactly what is sent.
Yes. The plugin is open source under Apache 2.0 and builds byte for byte from source, so you can check exactly what it sends. It holds no rules: those stay on our server. View on GitHub
Your agent’s security advisor, one prompt away.
- Works in the coding agent you already use
- Reviews start on your next turn
- ISO 27001 certified
