Veria Labs

Everything in one feed

Every finding across your repos in one live view, ranked by what's actually exploitable.

In your workflow

Findings post to Slack and file to Linear, with status kept in sync.

Verified exploits

Every finding comes with a working exploit, run against your staging environment.

Automatic fixes

Veria writes the patch and opens the PR with checks passing; all that's left is review and merge.

Security review on every PR

Veria catches the flaws that slip past human review, and leaves the fix inline.

Merged feat(devex): distributed owners.yaml ownership with a single resolver #68872

veria-ai bot reviewed 1d ago

‎.github/scripts/assign-reviewers.js‎ +1 −1

@@ -59,3 +59,3 @@ function resolveOwners(filenames) {

59 59 "import sys, runpy; sys.path.insert(0, 'tools/owners'); runpy.run_module('posthog_owners', run_name='__main__')"

60 - const result = spawnSync(python, ['-c', launcher], {

60 + const result = spawnSync(python, ['-I', '-c', launcher], {

61 61 input: filenames.join('\n'),

high Repository-root module shadowing

The python -c invocation retains the current repository directory on sys.path immediately after tools/owners. An attacker can land an unprotected root-level yaml.py or yaml/__init__.py, then have it imported instead of PyYAML by a future pull_request_target run and read the app token from GITHUB_TOKEN. Run Python in isolated mode so the working directory and environment-provided import paths are excluded.

View on GitHub ↗

Merged feat: add Infisical OAuth 2.0 support #6772

veria-ai bot reviewed 1mo ago

‎backend/src/server/plugins/auth/verify-auth.ts‎ +1 −1

@@ -26,3 +26,3 @@ const isAccessAllowed = authStrategies.some((strategy) => {

26 26 // from delegated access, omit AuthMode.JWT or check req.auth.oauthClientId in the handler.

27 - if (strategy === AuthMode.JWT && req.auth.authMode === AuthMode.OAUTH) return true;

27 + if (strategy === AuthMode.JWT && req.auth.authMode === AuthMode.OAUTH) return false;

28 28 return false;

high OAuth tokens bypass scopes on JWT-only account routes

AuthMode.OAUTH is accepted anywhere a route lists AuthMode.JWT, but OAuth scopes are only enforced when handlers later build org/project permissions. A delegated token with empty or unrelated scopes can still call JWT-only user routes such as GET /api/v1/user/me/totp to read TOTP recovery codes, DELETE /api/v1/user/me/totp to remove MFA, or session revocation endpoints because those handlers trust authentication alone.

View on GitHub ↗

Merged feat(mcp/ui): expose true_passthrough and oauth_delegate auth types with a no-auth warning #32414

veria-ai bot reviewed 1w ago

‎litellm/proxy/_experimental/mcp_server/server.py‎ +1 −1

@@ -125,3 +125,3 @@ def _redact_mcp_resource_url(url: Optional[str]) -> Optional[str]:

125 125 netloc = f"{parts.hostname}:{parts.port}" if parts.port else parts.hostname

126 - return urlunsplit((parts.scheme, netloc, parts.path, "", "")) or None

126 + return urlunsplit((parts.scheme, netloc, "", "", "")) or None

127 127

high Path credentials are written to spend logs

MCP URLs can carry credentials in the path, including formats such as /mcp/<token>. Because this value is persisted in mcp_tool_call_metadata, a user can invoke an allowed tool, read the resulting request log, and recover the upstream credential. Log only the origin unless there is a credential-aware path sanitizer.

View on GitHub ↗

Real reviews on public open-source PRs.

Leave nothing to chance.

Veria runs where mistakes are expensive

See it work

Everything in one feed

In your workflow

Verified exploits

Automatic fixes

Security review on every PR

Research

Know before an attacker does.