tech-article-creator
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted external sources, including web search results and user-provided reference materials (GitHub repositories, URLs), which is then processed by writing and reviewing agents. The lack of robust sanitization or specific boundary markers for this external data introduces a potential attack surface where instructions embedded in external content could influence the agent's behavior.
- Ingestion points: Data enters the context during 'Stage 1: Topic Selection' via WebSearch and 'Stage 3: Material Collection' via subagents gathering content from the web and user-referenced files, as described in
SKILL.mdandreferences/material-collection-guide.md. - Boundary markers: The prompt templates defined in
references/review-prompt-templates.mddo not utilize explicit delimiters or 'ignore instructions' warnings when passing gathered materials to the reviewing and writing models. - Capability inventory: The skill possesses capabilities for file system writes (article generation), subagent orchestration (Agent Team), and external platform publishing via auxiliary skills for Blog, WeChat, and X (Twitter).
- Sanitization: While the
SKILL.mdandconfig.yamloutline a 'redaction strategy' for sensitive project information, there is no evidence of filtering or escaping logic designed to neutralize potential prompt injection attacks within the collected materials.
Audit Metadata