discord-post-news
Warn
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to run a shell command using
curl. It dynamically constructs the command string by inserting content read from a local file, which could lead to command injection if the content is not handled securely by the underlying shell. - [DATA_EXFILTRATION]: The skill accesses files within the
memory/approval/directory and transmits the extracted information to an external network endpoint (a Discord webhook). This pattern of reading local data and sending it off-system is a primary vector for data exfiltration. - [PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface by ingesting data from files that might be influenced by external sources. Evidence: 1. Ingestion points:
memory/approval/ai-news-*-newsletter.md. 2. Boundary markers: None specified to separate data from instructions. 3. Capability inventory: Execution ofcurlvia the system shell. 4. Sanitization: The instructions mention JSON escaping, which mitigates some payload risks but does not prevent logic-based injection.
Audit Metadata