lov-translation-review
Warn
Audited by Gen Agent Trust Hub on Aug 24, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The workflow in
SKILL.mdinstructs the agent to process.docxfiles by executing a shell command:pandoc "<chinese_file>.docx" -t plain --wrap=none. The variables<chinese_file>and<english_file>are derived from user-provided filenames. Without proper sanitization, an attacker could provide a filename containing shell metacharacters (e.g.,$(command)) to achieve arbitrary command execution within the agent's environment. - [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it ingests and processes the content of untrusted Chinese and English documents provided by the user. \n
- Ingestion points: Document reading step in
SKILL.md(Step 2) usingpandocor direct reading. \n - Boundary markers: The instructions do not define boundary markers or instruct the agent to ignore instructions embedded within the user-provided text. \n
- Capability inventory: The agent has the ability to execute shell commands (via the
pandocinstruction) and write files to the local system (Step 6). \n - Sanitization: There is no evidence of content sanitization or validation of the input files or filenames to prevent malicious instructions from influencing the agent's behavior.
Audit Metadata