embedded-project-loop
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script
scripts/run_arduino_evals.pyto verify the integrity of the project's state files. While this is a standard development practice for automated testing, the execution of local scripts constitutes a command execution surface. - [INDIRECT_PROMPT_INJECTION]: The skill reads and interprets data from various state files located in the
.arduino/directory to determine its next actions. This creates an attack surface where maliciously crafted local files could influence the agent's behavior. - Ingestion points: Multiple files in the
.arduino/directory, includinggoal.md,next-todo.md,board-profile.md,experiment-log.jsonl, andresume-token.md. - Boundary markers: No explicit delimiters or instructions are provided to the agent to help it distinguish between trusted and untrusted content within these files.
- Capability inventory: The agent has the capability to execute shell commands (specifically
python3) and read/write to the local filesystem. - Sanitization: There is no evidence of explicit sanitization, validation, or escaping of the content read from these state files before it is used to influence the agent's logic.
Audit Metadata