lastXdays
Pass
Audited by Gen Agent Trust Hub on Apr 7, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection surface. The skill ingests untrusted content from Reddit and X through API calls and presents this data to the agent, which could contain embedded instructions. It also employs tool output poisoning by printing direct instructions to the agent (e.g., 'Claude: Use your WebSearch tool...') to manage multi-step research flows.
- Ingestion points: Untrusted content from Reddit and X APIs entering the agent context via the research script output.
- Boundary markers: None implemented in the provided orchestrator or rendering logic to separate untrusted data from instructions.
- Capability inventory: The skill environment allows access to
Bash,Read,Write,AskUserQuestion, andWebSearchtools. - Sanitization: No evidence of content sanitization or instruction filtering for fetched API data was found in the provided files.
Audit Metadata