lynx-devtool
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill's main binary (
bin.mjs) usesnpxto download and execute theagent-lynxpackage from the NPM registry at runtime. TheSKILL.mdalso instructs the agent to usepnpxorpnpm addto install dependencies in temporary directories. - [COMMAND_EXECUTION]: The skill is built entirely around executing the
agent-lynxCLI. It enables a wide range of actions including device discovery, process management, and interaction with host applications on Android, iOS, and Desktop platforms via shell commands. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external device sources that could contain malicious instructions meant to influence the agent's behavior.
- Ingestion points: Device console logs (
get-console.md), UI element snapshots and text content (snapshot.md), and debug metadata (get-source-info.md). - Boundary markers: Absent; the tool returns raw CLI output to the agent context without explicit delimiters or instructions to disregard embedded commands.
- Capability inventory: The agent has the ability to execute shell commands via
agent-lynxand evaluate JavaScript on the target device. - Sanitization: No sanitization, escaping, or filtering of device-originating logs or UI text is performed before the data is presented to the agent.
- [DYNAMIC_EXECUTION]: The skill enables the execution of dynamic and reflective content.
- The
evaluatecommand allows the agent to run arbitrary JavaScript within the Lynx Virtual Machine on the connected device. - The
App.CallStaticVoidMethodmethod utilizes reflection to invoke static methods on the host application side. - The
bin.mjsscript dynamically assembles annpxcommand to run a versioned package based on the skill's local metadata.
Audit Metadata