skills/megastep/codex-skills/blog/Gen Agent Trust Hub

blog

Pass

Audited by Gen Agent Trust Hub on Feb 26, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process and analyze blog content, which introduces a surface for indirect prompt injection from external sources. \n
  • Ingestion points: The scripts/analyze_blog.py script reads markdown and HTML content from the local filesystem. The blog-researcher agent uses WebSearch to gather data from the web. \n
  • Boundary markers: The skill does not implement specific delimiters or 'ignore' instructions to isolate processed content from the agent's internal reasoning. \n
  • Capability inventory: The agent can perform file reads, web searches, and initiate complex sub-agent workflows using the spawn_agent command. \n
  • Sanitization: There is no evidence of sanitization or filtering for embedded instructions in the ingested blog content or research data. \n- [COMMAND_EXECUTION]: The skill includes a local script and utilizes agent orchestration for its core functionality. \n
  • Evidence: The file scripts/analyze_blog.py is a standalone utility for auditing content quality. The SKILL.md file describes an orchestration logic that involves spawning sub-agents (explorer, worker) to perform discovery and parallel tasks.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 26, 2026, 01:29 AM