In sandboxed runs, native prompt image auto-load did not honor tools.fs.workspaceOnly=true.
This optional hardening setting is not enabled by default. When operators enabled it, prompt text could still reference mounted out-of-workspace image paths (for example /agent/secret.png) and load those image bytes for vision-capable model input.
openclaw (npm)2026.2.23<= 2026.2.232026.2.24This issue required all of the following:
- sandbox mode enabled,
- tools.fs.workspaceOnly=true configured,
- an out-of-workspace mount path reachable from the sandbox (for example /agent),
- vision-capable model path active for native prompt image loading.
Native prompt image ingestion (detectAndLoadPromptImages / loadImageFromRef) resolved and read sandbox paths but did not apply the same workspace-root assertion used by file tools when tools.fs.workspaceOnly was set.
370d115549c0dadace0902775eea0d5094aedfdcpnpm checkpnpm exec vitest run --config vitest.gateway.config.tspnpm test:fastpatched_versions is pre-set to the planned next release (2026.2.24) so once npm release is available, this advisory only needs publish action.
OpenClaw thanks @tdjackey for reporting.
openclaw@2026.2.24 is published on npm and contains the fix commit(s) listed above. This advisory now marks >= 2026.2.24 as patched.
{
"github_reviewed_at": "2026-03-03T19:08:08Z",
"github_reviewed": true,
"cwe_ids": [
"CWE-200",
"CWE-284"
],
"nvd_published_at": null,
"severity": "HIGH"
}