agently-design
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references a research paper on the well-known repository arXiv.org (LongWriter) as a methodological reference for long-form generation strategies.
- [INDIRECT_PROMPT_INJECTION]: As an auditing and design skill, it is intended to ingest and analyze untrusted data such as user-provided design documents, execution logs, and trial evidence. The skill includes robust internal instructions for managing these boundaries, including:
- Ingestion points: Audit evidence, developer questions, and observed execution traces.
- Boundary markers: Instructions mandate the use of explicit delimiters, stable schemas, and the separation of 'hot' context from 'cold' evidence.
- Capability inventory: The skill primarily performs architectural analysis and routes actual implementation or execution to other dedicated skills (e.g., agently-runtime).
- Sanitization: Strong emphasis on the redaction of secrets, authentication headers, and PII before logging, as well as deterministic validation of all model-returned keys and enums.
Audit Metadata