[Eclipse Theia] Indirect Prompt Injection via Adversarial Workspace File and Directory Names in AI Chat
Eclipse Foundation Eclipse Theia
Affected versions
| Product | Affected | Fixed in |
|---|---|---|
| @theia/ai-chat-ui npm |
|
|
| @theia/ai-chat npm |
|
|
| @theia/ai-claude-code npm |
Eclipse Foundation Eclipse Theia
| Product | Affected | Fixed in |
|---|---|---|
| @theia/ai-chat-ui npm |
|
|
| @theia/ai-chat npm |
|
|
| @theia/ai-claude-code npm |
< 1.71.0
|
| @theia/ai-code-completion npm |
|
|
| @theia/ai-core npm |
|
|
| @theia/ai-editor npm |
|
|
| @theia/ai-ide npm |
|
|
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. An attacker could craft a malicious repository with adversarial directory or file names that, when analyzed by the AI agent, would cause the agent to follow attacker-controlled instructions (indirect prompt injection). Combined with other AI chat features available in untrusted workspaces, this enabled attack chains leading to data exfiltration via Markdown image rendering or arbitrary command execution via task definitions.
From the CVE record · eclipse
Before version 1.71.0, Eclipse Theia’s AI chat agent treated workspace file and directory names as prompt context without separating them from instructions. If the agent analyzes an attacker-crafted repository, malicious names can make it follow attacker-controlled instructions. In untrusted workspaces, this can combine with other AI chat features to exfiltrate data through Markdown image rendering or execute commands through task definitions.
Teams running Eclipse Theia with @theia/ai-chat-ui, @theia/ai-chat, @theia/ai-claude-code, @theia/ai-code-completion, @theia/ai-core, @theia/ai-editor, or @theia/ai-ide versions before 1.71.0 should check their installations.