CVE-2026-56696: medium-severity vulnerability in HKUDS OpenHarness
OpenHarness - Prompt Injection via /issue and /pr_comments Slash Commands
Published · Updated
E-mail alert when it lands in KEV, gets a public PoC or its score goes up.
No sign of exploitation. No public exploitation artifact known so far.
OpenHarness allows attackers to inject malicious commands through chat slash commands that aren't properly protected, letting them secretly add harmful content to project files that gets inserted into system prompts and changes how the local agent behaves.
CWE-862 (Missing Authorization) vulnerability in /issue and /pr_comments slash commands lacks remote_invocable=False protection, enabling unauthenticated remote channel senders to perform arbitrary writes to .openharness/issue.md and .openharness/pr_comments.md. Injected Markdown persists in runtime system prompts, allowing persistent prompt injection attacks that alter local agent behavior without detection.
In the same product, most dangerous first.