social-intelligence

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process unstructured data from external, attacker-controllable sources such as Reddit comments, Hacker News posts, and YouTube transcripts. This creates a surface for indirect prompt injection attacks if the agent processes this data without sufficient isolation.
  • Ingestion points: Scripts in guides/community-monitoring-guide.md and payloads.md fetch content from public social media APIs.
  • Boundary markers: The methodology and scripts do not explicitly define boundary markers or 'ignore embedded instructions' warnings for the ingested content.
  • Capability inventory: The skill utilizes Bash (via CLI tools), WebFetch, and Write capabilities across various files.
  • Sanitization: There is basic filtering like text slicing in scripts, but no robust prompt injection sanitization or escaping of external content.
  • [DYNAMIC_EXECUTION]: Several guide scripts utilize subprocess.run to execute external CLI tools such as snscrape, exiftool, dig, and yt-dlp with arguments derived from external data or user-defined keywords.
  • Evidence: guides/community-monitoring-guide.md (monitor_x function), guides/target-profiling-guide.md (correlate_identity function), and various shell pipelines in payloads.md.
  • [EXTERNAL_DOWNLOADS]: The skill references and provides instructions to download several well-known security and OSINT tools from GitHub repositories.
  • Evidence: Fetches tools such as Sherlock, yt-dlp, and SpiderFoot from their respective official repositories on GitHub, which is a well-known service for hosting such tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 04:36 PM