arcs-dev-tools
Pass
Audited by Gen Agent Trust Hub on Mar 11, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses
git cloneandgit submodule updateto download the ARCS SDK from a repository (defaulting to a LISTENAI vendor domain). - [COMMAND_EXECUTION]: Executes several shell commands for environment management, including
chmod +xto enable execution of thecskburnbinary andfuserto manage serial port access. - [REMOTE_CODE_EXECUTION]: Runs shell scripts (
prepare_listenai_tools.sh,prepare_toolchain.sh, andbuild.sh) fetched from the repository. This is part of the standard installation and build process for this toolchain. - [PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface by reading raw serial logs from hardware and providing them to the agent.
- Ingestion points:
serial_read.pyreads from/dev/ttyACM*and/dev/ttyUSB*devices. - Boundary markers: No explicit markers or delimiters are used to wrap serial output in
SKILL.md. - Capability inventory: Includes the ability to clone repositories, execute bash scripts, and flash binaries (referenced in
SKILL.md). - Sanitization: Serial data in
serial_read.pyis decoded as UTF-8 with character replacement but is not sanitized for potential embedded instructions.
Audit Metadata