social-content
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill runs a local Python script (
brand_context.py) to initialize brand-specific settings. This execution is confined to the local filesystem and is a standard part of the skill's setup process. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external social media profiles. It mitigates injection risks by using a structured extraction protocol that captures only abstract structural patterns (e.g., archetypes and short 10-word fragments) instead of full post content, preventing the storage or execution of hidden instructions.
- Ingestion points: Scrapes content from public X and LinkedIn accounts using APIs and web search tools in
SKILL.md(Phase A). - Boundary markers: Hard coding instructions to extract 'shapes, never sentences' and limiting hook fragments to 10 words maximum.
- Capability inventory: File system writing (
social-canon.md), network API access, and local script execution. - Sanitization: High-level abstraction into a structural taxonomy which filters out potential natural language attacks.
- [EXTERNAL_DOWNLOADS]: The skill connects to third-party services like X API, Firecrawl, Zernio, and Buffer to perform its primary functions of research and scheduling. These are legitimate integrations configured via user-provided credentials.
Audit Metadata