Public AI agent scan brief

OpenHands/OpenHands AI agent security scan

Partial no-execution triage for the public OpenHands repo. This is a heuristic scan of 90 selected public text files, not a commissioned audit, vulnerability disclosure, or security certification. These are not confirmed vulnerabilities.

Rerun public scan Coding agent audit path Start audit from Radar

RepoOpenHands/OpenHands

Score68/100 heuristic

Snapshot90 selected public text files

Signals1 high / 1 medium / 1 low

Top findings

Coding agents need clear write, shell, and browser controls

The scan snapshot saw 77,768 stars and review signals around write actions, remote service exposure, and credential paths with some redaction evidence. A paid review would confirm tool safety annotations, workspace boundaries, auth middleware, and tenant assumptions.

HighWrite actions detected without obvious tool safety annotations.

MedRemote listener needs explicit exposure policy.

LowCredential paths detected; redaction appears present.

Evidence slice

Files that triggered review signals

Write-action signals appeared around `.agents/skills/cross-repo-testing/SKILL.md`, `enterprise/tests/unit/test_saas_server.py`, frontend API tests, and `frontend/__tests__/utils/browser-tab.test.ts`.

Remote and auth-related signals appeared near `enterprise/tests/unit/server/test_constants.py`, `enterprise/tests/unit/test_auth_middleware.py`, `frontend/__tests__/api/v1-conversation-service.test.ts`, `frontend/__tests__/api/v1-git-service.test.ts`, and `frontend/__tests__/routes/vscode-tab.test.tsx`.

Confirm which write tools are read-only, reversible, confirmed, or destructive.

Validate remote runtime exposure, authentication, and workspace isolation.

Test secrets, API keys, browser tabs, git operations, and conversation state boundaries.

Paid handoff

Convert this scan into a fixed-scope review

Confirm the current coding-agent surface, write actions, shell access, and hosted mode.
Review auth middleware, browser tab, git, filesystem, and API proxy boundaries.
Validate redaction, approval gates, default-deny behavior, and regression tests.
Pay USD $1,000 only after written scope acceptance.

Start AI agent audit Open quote Back to AI Radar