wjs-distilling-style
Pass
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill includes documentation for using
curlcommands to synchronize style cards with the author's platform (jianshuo.dev). These operations are transparent, target a vendor-owned domain, and require a user-provided token. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data in the form of sample articles. While this presents a theoretical surface for indirect prompt injection, the risk is mitigated by the structured distillation process and the use of sub-agents with narrow instructions.
- Ingestion points: User-provided sample articles stored in the
~/code/style-cards/<author-slug>/samples/directory. - Boundary markers: The analysis workflow uses structured 9-axis templates and dedicated sub-agent prompts (
references/judge-prompt.md) to isolate the analytical task from general instruction following. - Capability inventory: The skill possesses file system access (read/write in
~/code/style-cards/) and network access via thecurlutility. - Sanitization: No explicit sanitization of article text is described, but the extraction-focused nature of the distillation process typically prevents the execution of embedded instructions.
Audit Metadata