article-factory
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The orchestrator executes a local shell script at
../catalog-deslop/references/lint.shto validate the draft's prose against specific stylistic constraints between the writer and reviewer stages. - [EXTERNAL_DOWNLOADS]: The research subagent is instructed to use web access to identify first-person practitioner sources (engineering blogs, postmortems, etc.) to build a factual foundation for the article.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the web (research) and user-provided inputs (briefs). While this constitutes an attack surface, the pipeline incorporates human-like review gates (Cold Reviewer) and automated refinement (Deslop) to manage content quality, and the agent's capabilities are focused on text generation.
- [DYNAMIC_EXECUTION]: The skill chains to another agent skill,
catalog-deslop, to perform final text normalization. This is a modular orchestration pattern common in multi-agent systems.
Audit Metadata