smart-router-skill
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- COMMAND_EXECUTION (HIGH): The skill orchestrates shell execution using user-provided strings. Specifically, the instruction to run
bash router.sh --character <your_choice>interpolates the user's response directly into a shell command. - Ingestion points: User selection prompt described in
SKILL.mdunder the 'Example Flow' section. - Boundary markers: Absent; the command-line pattern lacks quoting or escaping for the variable.
- Capability inventory: Access to the local shell environment via
bashand sub-scripts in thescripts/directory. - Sanitization: Absent; the instructions do not explicitly direct the agent to validate or whitelist the input against the allowed characters ('yoda', 'stark', 'sherlock').
- PROMPT_INJECTION (MEDIUM): The interactive nature of the skill allows a user to 'inject' malicious payloads into the system by providing inputs containing shell metacharacters (e.g.,
;,&&, or|) when the agent asks for a character choice. This could lead to arbitrary code execution on the host machine.
Recommendations
- AI detected serious security threats
Audit Metadata