probe-rs
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill wraps the
probe-rsandarm-none-eabi-gdbCLI tools using the Pythonsubprocessmodule to perform hardware operations. - Evidence: Subprocess calls are implemented in
scripts/probe_rs_exec.py,scripts/probe_rs_gdb.py, andscripts/probe_rs_rtt.py. - Analysis: The implementation uses list-based arguments for all subprocess calls, which is a secure practice to prevent shell injection vulnerabilities. The skill correctly handles tool paths from configuration or system PATH.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data originating from hardware devices, such as RTT logs and memory contents, which constitutes a potential attack surface.
- Ingestion points:
scripts/probe_rs_rtt.pyreads RTT streams from the target hardware;scripts/probe_rs_exec.pyparses memory read results. - Boundary markers: Data is parsed and encapsulated in structured JSON objects before being returned to the agent context, providing clear delimitation.
- Capability inventory: The skill has capabilities to read/write memory and flash firmware, which are its primary intended functions for hardware development.
- Sanitization: Output is processed via regex-based parsers to extract relevant information, minimizing the risk of raw device output influencing agent behavior.
- [DATA_EXPOSURE]: The skill manages local state and configuration within the project workspace to maintain session context.
- Evidence:
scripts/probe_rs_runtime.pyinteracts with.embeddedskills/state/probe-rs.jsonand.embeddedskills/config.json. - Analysis: This is standard behavior for maintaining project-specific settings and session history. No hardcoded credentials or sensitive file exfiltration patterns were detected.
Audit Metadata