social-content
Pass
Audited by Gen Agent Trust Hub on Feb 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill identifies a workflow for 'Reverse Engineering Viral Content' in the file 'references/reverse-engineering.md' which presents a vulnerability to indirect prompt injection.
- Ingestion points: The instructions guide the agent to 'SCRAPE' and 'Collect 500-1000+ posts' from external platforms like LinkedIn and Twitter/X for analysis.
- Boundary markers: There are no specified delimiters or 'ignore embedded instructions' warnings for the data ingested from these external social media posts.
- Capability inventory: The agent is tasked with qualitative analysis including 'What emotional triggers appear' and 'Extracting patterns,' which requires the LLM to process and follow patterns found in untrusted content.
- Sanitization: The skill lacks any steps for sanitizing, filtering, or validating the text content gathered from external creators before it is processed by the agent context.
Audit Metadata