[Eclipse Theia] Indirect Prompt Injection via Adversarial Workspace File and Directory Names in AI Chat - @theia/ai-claude-code

Description

In Eclipse Theia versions prior to 1.71.0, the AI chat agent processed workspace file and directory names as part of its prompt context without distinguishing them from system instructions.

Recommendation

Update the @theia/ai-claude-code package to the latest compatible version. Followings are version details:

• Affected version(s): < 1.71.0
• Patched version(s): 1.71.0

References

• GHSA-3jww-hxqj-wfq2
• gitlab.eclipse.org
• CVE-2026-44688
• CWE-829
• CAPEC-310
• OWASP 2021-A6
• OWASP 2021-A8

Related Issues

• [Eclipse Theia] Indirect Prompt Injection via Adversarial Workspace File and Directory Names in AI Chat - @theia/ai-code-completion - CVE-2026-44688
• [Eclipse Theia] Indirect Prompt Injection via Adversarial Workspace File and Directory Names in AI Chat - @theia/ai-chat-ui - CVE-2026-44688
• [Eclipse Theia] Indirect Prompt Injection via Adversarial Workspace File and Directory Names in AI Chat - @theia/ai-chat - CVE-2026-44688
• [Eclipse Theia] Indirect Prompt Injection via Adversarial Workspace File and Directory Names in AI Chat - @theia/ai-core - CVE-2026-44688

You might also like:

How do hackers hack websites?

10 Secure Coding Best Practices to Follow in Every Project

Update Cheat Sheet for Developers

Tags:

npm

@theia/ai-claude-code