linkedin-content

Pass

Audited by Gen Agent Trust Hub on Sep 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes a scoped Bash tool constrained to belt * commands. This allows the agent to interact with the inference.sh platform ecosystem for tasks like research (tavily/search-assistant), image generation (falai/flux-dev-lora), and social media posting (x/post-create).
  • [EXTERNAL_DOWNLOADS]: The instructions prompt the installation of the belt-sh/cli via npx and suggest adding supplemental skills from the inference-sh repository. These downloads are directly related to the skill's primary functionality of content automation.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface where it ingests external data from the tavily/search-assistant application. Adversarial content within search results could theoretically attempt to influence the agent's output. However, this is a standard operational risk for research-oriented skills and is mitigated by the fact that the output is intended for human review before posting.
  • Ingestion points: External search queries via tavily/search-assistant (SKILL.md).
  • Boundary markers: None explicitly defined in the example command.
  • Capability inventory: Restrictive Bash access limited to the belt CLI tool.
  • Sanitization: Standard LLM processing of search results.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 2, 2026, 02:18 PM
Security Audit — agent-trust-hub — linkedin-content