wordpress
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill demonstrates code patterns that ingest untrusted data and process it via AI connectors without adequate safety boundaries.
- Ingestion points: The
wp_ai_client_promptcalls in the 'WordPress 7.0 AI Connector Usage' and 'Abilities API Registration' sections ingest content from$post->post_content(SKILL.md). - Boundary markers: Absent. User data is concatenated directly to instructions (e.g.,
'Summarize in 2 sentences: ' . substr($content, 0, 1000)) without delimiters or instructions to ignore embedded commands. - Capability inventory: The workflow bundle orchestrates high-privilege skills such as
@deployment-engineer,@security-auditor, and@wordpress-penetration-testing. The specific code examples include database write capabilities (wp_update_post). - Sanitization: The code uses
strip_tags()andsubstr(), which remove HTML and limit length but do not filter out plain-text natural language instructions designed to hijack the AI's behavior.
Audit Metadata