nmbrs
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents a Proxy API escape hatch that allows the agent to interact with arbitrary downstream endpoints via the
x-apideck-downstream-urlheader. This introduces a surface for indirect prompt injection, as the agent could be manipulated into sending data to a malicious destination if the target URL is derived from untrusted user input without validation.\n - Ingestion points: The
x-apideck-downstream-urlheader used in Proxy API examples inSKILL.mdcould be influenced by external input.\n - Boundary markers: There are no explicit instructions or delimiters for the agent to treat the downstream URL as a restricted or sensitive value.\n
- Capability inventory: The skill includes network interaction capabilities using
curlto reach the Apideck Unify proxy service.\n - Sanitization: No sanitization, validation, or allow-listing mechanism for the downstream URL is described in the skill instructions.
Audit Metadata