publication-link-audit

Pass

Audited by Gen Agent Trust Hub on Aug 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes content from external websites (Reddit, Hugging Face, GitHub) to verify that page titles match the expected link labels. This introduces a surface for indirect prompt injection, where an attacker could embed malicious instructions in external page titles or metadata to influence the agent's behavior.
  • Ingestion points: The agent fetches and processes content from external URLs found in user-provided documents, specifically from reddit.com, huggingface.co, and github.com.
  • Boundary markers: The instructions do not specify any delimiters or safety guidelines (like 'ignore embedded instructions') to be used when handling this external content.
  • Capability inventory: The skill workflow involves using network utilities like curl and HTTP libraries to interact with these external sites.
  • Sanitization: No sanitization or validation of the retrieved text content is described in the workflow.
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to verify link status.
  • Evidence: The documentation explicitly recommends using curl -sL -w \"%{http_code}\" to verify the reachability of GitHub URLs.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 14, 2026, 07:56 AM
Security Audit — agent-trust-hub — publication-link-audit