content-analysis-report
Pass
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill proactively addresses the risk of indirect prompt injection by including a clear instructional block warning the agent that scraped content is untrusted data and not instructions. It specifically directs the agent to treat data from the database (captions, transcripts) as material for analysis only and to never follow instructions found within that content.
- Ingestion points: Data enters the context from the Supabase
contenttable (SKILL.md). - Boundary markers: Present; uses a high-visibility blockquote warning at the start of the instructions.
- Capability inventory: The skill is limited to reading data and writing a report to a local markdown file.
- Sanitization: Explicitly instructs the agent to quote addressing text and stop if content appears to address the AI directly.
- [DATA_EXFILTRATION]: No evidence of unauthorized data exfiltration was found. The skill queries a Supabase database using a configuration placeholder (
YOUR_SUPABASE_PROJECT_ID) and writes the output to a local project directory. It does not perform any outbound network requests to unknown third-party services. - [COMMAND_EXECUTION]: The skill uses structured SQL queries to interact with the database. There is no usage of arbitrary shell commands, subprocess spawning, or dynamic execution patterns that could be exploited.
Audit Metadata