Openclaw cli: OpenClaw macOS deep link confirmation truncation can conceal executed agent message (CVE-2026-26320)

The OpenClaw macOS desktop client registers the openclaw:// URL scheme and shows a confirmation dialog for agent deep links without an unattended key. Previously, this dialog only displayed the first 240 characters of the message, but executed the entire message after user approval. An attacker could exploit this by padding the message with whitespace to hide malicious payloads beyond the visible preview, increasing the likelihood of user approval of unintended commands. The vulnerability affects versions from 2026.2.6 up to and including 2026.2.13 and is resolved in 2026.2.14 by enforcing strict message length limits and ignoring certain parameters unless a valid unattended key is provided.

If a user approves a crafted deep link prompt, the OpenClaw agent may execute arbitrary commands depending on the user's configured tool approvals or allowlists. This vulnerability relies on social engineering to mislead users about the actual command being executed, potentially leading to unauthorized actions on the affected system.

A patch is available in OpenClaw macOS desktop client version 2026.2.14. Users should upgrade to this version or later. Until patched, users should not approve unexpected "Run OpenClaw agent?" prompts from untrusted sources and should use unattended deep links only with a valid key for trusted automations. The fix enforces message length limits and restricts execution parameters without a valid unattended key.