GHSA-6c4r-g249-wv3c: OpenClaw: Sandboxed session spawn could expose the real workspace path to child prompts

A sandbox is meant to hide the host layout from the agent inside it. In affected versions, a child session spawned from a sandboxed parent got the host workspace path forwarded into its prompt. The child model could then reveal that path in its output or use it in later steps. The advisory says it could also expose related memory context. It does not say what that context holds or which function builds the prompt.

The issue is information exposure (CWE-668), not code execution. The CVSS 4.0 vector is network, low complexity, low privileges, no user interaction, with low confidentiality impact only. An attacker needs a session on the Gateway, or influence over a prompt that causes a spawn from a sandboxed parent. They learn a host path, which helps with follow-on attacks such as path-based escapes. The impact depends on how sensitive your host paths and memory content are.

The advisory leaves OpenClaw's trusted-operator model unchanged.

Affected: npm openclaw up to and including 2026.4.25. You are exposed if you use sandboxed sessions that can spawn child sessions.

npm ls -g openclaw

Drop -g for a project install. Check your agent configuration for sandboxed workspaces that allow session spawn.

Update to openclaw 2026.4.26, the first stable patched version.

npm install -g [email protected]

Until you can update, avoid spawning child sessions from sensitive sandboxed workspaces. The advisory also suggests narrow channel and tool allowlists, not sharing one Gateway between mutually untrusted users, and disabling the feature if you do not need it.

The advisory gives no log signature. Inspect the prompts and transcripts of child sessions spawned from sandboxed parents. Search them for your real host workspace path, such as the home directory or checkout location, rather than the sandbox path. Check model outputs and downstream messages for the same path, since the child may have repeated it.

Summary

Sandboxed session spawn could expose the real workspace path to child prompts. In affected versions, a child session spawned from a sandboxed parent could forward the host workspace path into the child session prompt.

This advisory is scoped to the named feature and configuration. It does not change OpenClaw's trusted-operator model: authenticated Gateway operators, installed plugins, and intentional local execution surfaces remain trusted unless a separate policy, approval, allowlist, sandbox, or auth boundary is crossed.

Impact

When the affected feature is enabled and reachable, this could reveal host workspace location or related memory context to the child model. Practical impact depends on the operator's configuration and whether lower-trust input can reach that path.

Patched Versions

The first stable patched version is 2026.4.26.

Mitigations

avoid spawning child sessions from sensitive sandboxed workspaces until patched. As general hardening, keep channel and tool allowlists narrow, avoid sharing one Gateway between mutually untrusted users, and disable the affected feature when it is not needed.