blockchain-solana
Pass
Audited by Gen Agent Trust Hub on Aug 29, 2026
Risk Level: SAFEOBFUSCATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [OBFUSCATION]: The files
references/blockchain-solana-ref-1.mdandreferences/blockchain-solana-ref-2.mdcontain more than 50 repetitions of the same markdown headers and Rust code blocks. This redundant content is characteristic of 'context stuffing', an adversarial technique designed to fill the agent's context window to displace safety guidelines or system instructions. - [INDIRECT_PROMPT_INJECTION]: The skill architecture is susceptible to indirect prompt injection due to its processing of untrusted user-supplied context for program synthesis without delimiters or sanitization.
- Ingestion points: Input variables defined in
SKILL.mdsuch as application type, cluster environment, account model design, and external dependencies. - Boundary markers: The instructions do not define any delimiters or specific framing to isolate user-provided data from the agent's task logic.
- Capability inventory: The skill is capable of generating complex blockchain program code (Rust/Anchor) and client integration tests (TypeScript/Web3.js) across all reference documents.
- Sanitization: No evidence of sanitization or validation protocols for the external context being interpolated into the code generation workflow.
Audit Metadata