Eclipse Foundation Theia vulnerabilities

In Eclipse Theia versions up to and including 1.73.1, the `@theia/filesystem` backend exposes HTTP file-download endpoints (`GET /file`, `GET /files/`, `PUT /files/`) that convert a client-supplied URI directly to a filesystem path and stream the file, without confining it to the workspace or any allow-listed root. In browser (non-Electron) deployments the connection token is enforced only on WebSocket upgrades; the HTTP middleware in `@theia/core` re-issues the cookie and calls `next()` without rejecting tokenless HTTP requests, so these endpoints are reachable without a valid token. As a result an unauthenticated client can read any file readable by the backend process, including files outside the opened workspace (for example `/etc/hosts`, SSH keys, or tokens). Electron mode uses a separate `ElectronSecurityToken` and is not affected via this path.

In Eclipse Theia versions up to and including 1.73.1, the `@theia/filesystem` backend binds `POST /file-upload` in every filesystem-enabled deployment. The handler takes an attacker-supplied absolute path from the multipart `uri` field and calls `fs.move(tmp, target, { overwrite: true })` with no workspace confinement and no authentication. In browser (non-Electron) deployments the connection token is enforced only on WebSocket upgrades; the HTTP middleware in `@theia/core` re-issues the cookie and calls `next()` without rejecting tokenless HTTP requests. Because `multipart/form-data` is a CORS-safelisted request type, a cross-origin web page can trigger the write with no preflight and no credentials, resulting in an unauthenticated arbitrary file write outside the workspace to any absolute path the backend process can write. This can escalate to remote code execution, for example by overwriting a startup-executed file such as `~/.bashrc`. Electron mode uses a separate `ElectronSecurityToken` and is not affected via this path.

In Eclipse Theia versions 1.66.0 and up until including 1.73.1, the `@theia/plugin-ext` backend exposes the `/hostedPlugin/:pluginId/:path(*)` HTTP endpoint, which resolves the requested file path with `path.resolve(localPath, filePath)` without verifying that the resolved path stays within the plugin's directory. An unauthenticated network attacker can send percent-encoded `../` sequences (`%2e%2e%2f`) that decode into the path parameter and escape the plugin directory, allowing arbitrary files readable by the Theia backend process to be retrieved. Plugin IDs are derived deterministically from a plugin's publisher and name, so built-in plugins serve as reliable anchors that require no prior knowledge of the target system.

In Eclipse Theia since version 1.26.0, the backend /services/request-service RPC accepts an attacker-controlled URL from any client connected to the standard /services messaging endpoint, performs the HTTP request server-side, and returns the full response body to the caller. Because the destination URL is neither validated nor allowlisted, a remote attacker with access to the Theia service connection can issue server-side HTTP requests to localhost or other backend-reachable hosts and read their responses, exposing internal administrative endpoints, cloud instance metadata services, and other resources that are intentionally outside the browser network boundary. The vulnerability affects deployments where the Theia service connection is reachable by untrusted users (for example, multi-tenant or publicly-reachable Theia deployments).

In affected versions of Eclipse Theia (1.8.1 and later), the browser backend exposes privileged terminal RPC over WebSocket (/services/shell-terminal, /services/terminals/:id) without service-level authentication. WebSocket origin validation in @theia/core is fail-open: connections are accepted when the Origin header is missing or when no THEIA_HOSTS allowlist is configured (the default). The Socket.IO integration additionally replaces the real Origin header with a client-supplied fix-origin header that an attacker can control or omit. As a result, a foreign-origin web page visited by a user with a running Theia instance can open the /services WebSocket namespace, invoke terminal creation, attach to the resulting terminal data channel, execute arbitrary OS commands, and read their output. This affects both local developer setups (drive-by attack) and hosted or tunneled deployments without strong external authentication. A fix is in development that enforces same-origin validation by default, removes trust in the fix-origin header, gates HTTP and WebSocket access on a SameSite=Strict; HttpOnly connection-token cookie, and sanitizes shell terminal creation options.

In Eclipse Theia versions prior to 1.71.0, files matching the pattern .prompts/*.prompttemplate in a workspace were automatically loaded and could override or extend the AI agent's system prompts. An attacker could craft a malicious repository containing prompt template files that, when the workspace was opened in Theia, replaced the AI's system instructions with attacker-controlled content (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.

In Eclipse Theia versions prior to 1.69.0, custom task definitions in workspace files (e.g. .theia/tasks.json, .vscode/tasks.json) could be executed without requiring workspace trust. An attacker could craft a malicious repository that, when cloned and opened in Theia, leads to execution of arbitrary commands with the user's privileges. In combination with AI chat features and a workspace .theia/settings.json that disabled tool confirmation, this could be triggered automatically by sending a message in the AI chat.

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.