publish-blog

Warn

Audited by Gen Agent Trust Hub on May 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes multiple Python scripts (content-scorer.py, brand-voice-scorer.py, approval-manager.py, and execution-tracker.py) during the publishing workflow. As these scripts are not included in the skill package, their contents cannot be verified for safety.
  • [DATA_EXFILTRATION]: The instructions require reading from the user's home directory (e.g., ~/.claude-marketing/brands/). This pattern of accessing application-specific configuration files outside the project scope can lead to the exposure of sensitive brand profiles or local settings.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted "Blog content" drafts that could contain malicious instructions.
  • Ingestion points: User-provided blog drafts (SKILL.md).
  • Boundary markers: No explicit delimiters or security warnings are used to separate the untrusted draft content from the agent's instructions.
  • Capability inventory: The skill has the ability to execute local scripts and interact with external CMS platforms via MCP servers.
  • Sanitization: The process includes quality and brand voice scoring, but lacks security-focused sanitization to prevent the agent from executing instructions embedded within the draft content.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 17, 2026, 09:32 PM
Security Audit — agent-trust-hub — publish-blog