wiki-room
Warn
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The instruction to run python3 harness/scripts/mem-rank.py retrieve "<câu hỏi>" creates a risk where unsanitized user input in the <câu hỏi> placeholder could lead to shell command injection if shell metacharacters are present.
- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface.
- Ingestion points: User input is ingested via the <câu hỏi> variable in SKILL.md.
- Boundary markers: None are used to delimit the user input when passed to the subagent or the Python script.
- Capability inventory: The skill executes shell commands via python3 and spawns subagents.
- Sanitization: No sanitization or validation of the user input is specified before interpolation.
Audit Metadata