browse
Pass
Audited by Gen Agent Trust Hub on Jun 24, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: Indirect prompt injection surface identified through untrusted data ingestion. * Ingestion points: Scrapes content from the Twitter (X) home timeline (x.com/home) as described in workflows/twitter-feed.md. * Boundary markers: Absent. The skill does not use specific delimiters or instructions to separate scraped tweet content from internal processing directives. * Capability inventory: Browser interaction via mcp__claude-in-chrome and file system access (read/write) to ~/CybosVault. * Sanitization: The workflow includes relevance filters and quality gates to categorize content, but lacks explicit sanitization to prevent adversarial instructions within tweets from influencing agent behavior.
- [DATA_EXFILTRATION]: Accesses sensitive local file paths within the user's private data vault. * The workflow reads account information from ~/CybosVault/private/context/identity.md. * Discovered topics and operation logs are saved to subdirectories within ~/CybosVault/private/.
Audit Metadata