feishu-paper-reviewer
Pass
Audited by Gen Agent Trust Hub on Mar 31, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: Indirect prompt injection vulnerability identified. The skill ingests untrusted data from external Feishu documents that could contain instructions meant to override agent behavior.
- Ingestion points: Document content is fetched via
lark-cli docs +fetchas described in SKILL.md. - Boundary markers: No delimiters or protective instructions are present to prevent the agent from obeying commands embedded within the fetched paper content.
- Capability inventory: The skill possesses write capabilities including document modification (
lark-cli docs +update) and adding comments (lark-cli drive +add-comment). - Sanitization: There is no evidence of sanitization or validation of the document content before it is processed by the LLM or used in CLI command parameters.
- [COMMAND_EXECUTION]: The skill uses a command-line interface (
lark-cli) to perform document operations. It constructs commands using data derived from the paper and model-generated feedback, which creates a risk of command injection if the underlying execution platform does not properly sanitize shell arguments.
Audit Metadata