CVE-2025-58371

CVE-2025-58371 affects Roo Code (versions ≤ 3.26.6). A GitHub workflow used unsanitized pull request metadata in a privileged context, enabling an attacker to craft input that caused Remote Code Execution (RCE) on the Actions runner. The runner’s broad permissions and access to repository secrets...

CVE-2025-53536

Roo Code (AI-powered autonomous coding agent) prior to version 3.22.6 is affected. If a victim had the Write auto-approved mode, an attacker who can submit prompts could write to VS Code settings files and trigger code execution. A concrete example is the php.validate.executablePath setting, wher...

CVE-2025-65946

Roo Code (AI-powered coding agent) had a validation error before version 3.26.7 that could cause it to automatically execute commands not on the allowed prefixes list. The issue has been patched in version 3.26.7. Affected CVE-2025-65946 entries from multiple feeds confirm the vulnerability and p...

CVE-2025-53098

Roo Code prior to version 3.20.3 stores the MCP configuration in .roo/mcp.json. The MCP config format allows executing arbitrary commands, enabling an attacker who can submit prompts (e.g., via prompt injection) and who has MCP enabled and auto-approve file writes turned on to inject a malicious ...

CVE-2025-58372

Roo Code CVE-2025-58372 affects versions ≤3.25.23 where certain VS Code workspace files (.code-workspace) aren’t protected like the .vscode folder. If auto-approve for file writes is enabled and prompts are manipulated (e.g., via prompt injection), an attacker could write malicious workspace sett...

CVE-2025-58374

Summary (CVE-2025-58374): Roo Code versions 3.25.23 and earlier allow an auto-approved npm install that can execute a repository’s postinstall script, enabling arbitrary code execution. Root cause: npm install is in the default auto-approve list, so malicious postinstall scripts run without user ...

CVE-2025-54377

CVE-2025-54377 affects Roo Code. In versions 3.23.18 and earlier, Roo Code fails to validate line breaks in command input, allowing multi-line command injection by smuggling additional commands beyond the first line during evaluation. The issue stems from missing parsing/validation logic for inpu...

CVE-2025-53097

Roo Code extension (pre-3.20.3) allowed read access via the search_files tool outside the VS Code workspace, enabling potential data exposure if an attacker injects prompts. The attacker could exfiltrate data by writing to a JSON schema when the schema-fetch feature is enabled by default, trigger...

CVE-2025-58370

Roo Code (AI-powered coding agent) versions prior to 3.26.0 contain a vulnerability in the command parsing logic where Bash parameter expansion and indirect references are not handled correctly. If prompts allow auto-approval of commands, an attacker who can influence prompts could cause the agen...

CVE-2025-58373

Roo Code (editor-integrated AI coding agent) versions 3.25.23 and earlier contain a symlink-based bypass of the .rooignore protections. An attacker with write access to the workspace could trick the extension into reading files that should be excluded (for example, .env or other configuration dat...

CVE-2026-30307

The Roo Code AI-based autonomous coding agent contains a critical OS command injection vulnerability within its command auto-approval module . The root cause is the use of fragile regular expressions to parse command structures, which fails to account for standard Shell command substitution (e.g....

CVE-2026-110202

Roo-Code versions up to and including 3.54.0 are vulnerable to a command approval bypass in the getCommandDecision() function. The root cause is redirect operator stripping , where attackers can auto-approve denied shell commands by prefixing them with a bare >& redirect. By using indirect pro...

CVE-2026-110205

Roo-Code versions through 3.54.0 are vulnerable to a command approval bypass in the parseCommandLine() function. The flaw occurs because the system discards command substitutions when they are placed adjacent to ＊ or ！ (glob characters). An attacker leveraging indirect prompt injection can craft ...

CVE-2026-110203

Roo-Code versions through 3.54.0 are vulnerable to a command approval bypass caused by a quote-masking regex flaw in the parseCommandLine() function. Attackers can utilize indirect prompt injection with escaped or single-quoted double quotes to conceal command separators behind allowlisted prefix...

CVE-2026-110204

Roo Code versions through 3.54.0 are vulnerable to a command approval bypass due to how the parseCommandLine() function handles ${ command; } substitutions as parameter expansions. Attackers can utilize indirect prompt injection to embed unauthorized shell commands within allowlisted commands, su...

CVE-2026-110206

Roo-Code versions up to and including 3.54.0 are vulnerable to a command approval bypass in the parseCommandLine() function. The root cause is an interpretation conflict where attackers can use command substitution at the command position to bypass the software's allowlist. By using indirect prom...