arete-programs
Pass
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses the
a4CLI to search for, explore, and install generated program SDKs and stacks from a remote registry (e.g.,a4 install program <program-ref>). These resources are provided by the vendor (aretea4) to facilitate protocol integration. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from the Arete program registry and knowledge base via CLI commands like
a4 knowanda4 exploreto inform instruction building and transaction preparation. - Ingestion points: Remote program descriptors and metadata fetched from the Arete registry.
- Boundary markers: The skill explicitly documents an "Authority Boundary" in
references/transaction-safety.md, separating account reading and building from signing and submission. - Capability inventory: The skill allows for local file installation (
a4 install), account state reads, and transaction execution through a connected wallet. - Sanitization: Generated SDK builders are designed to "fail closed" on unknown parameters or missing required values, providing a layer of protection against malformed descriptor data.
Audit Metadata