skills/sendaifun/skills/switchboard/Gen Agent Trust Hub

switchboard

Warn

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONMETADATA_POISONINGINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DATA_EXFILTRATION]: Multiple script files and templates in the skill are designed to read the default Solana wallet file located at ~/.config/solana/id.json. This path is accessed via fs.readFileSync in examples/setup/example.ts, examples/feeds/oracle-quote.ts, examples/feeds/pull-feed.ts, examples/feeds/read-feed.ts, templates/setup.ts, and examples/randomness/vrf-example.ts. While standard for local development of blockchain applications, it represents access to sensitive private keys.
  • [METADATA_POISONING]: There is an inconsistency in the skill's identity as the creator is stated as raunit-dev in the frontmatter, while the platform identifies the author as sendaifun.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides an integration framework for processing external data from oracles and APIs, which introduces a potential attack surface.
  • Ingestion points: Oracle feeds and price updates fetched through the CrossbarClient and SwitchboardSurge WebSocket stream.
  • Capability inventory: The skill can build and send Solana transactions, perform network requests, and execute filesystem operations.
  • Boundary markers: No specific delimiters or warnings are used in the prompt logic to differentiate between oracle data and instructions.
  • Sanitization: No clear sanitization steps for external feed data were found in the examples.
  • [EXTERNAL_DOWNLOADS]: The skill instructions include downloading official SDK packages and cloning repositories from the Switchboard and Coral organizations. These are recognized services within the Solana development community.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 14, 2026, 10:07 PM
Security Audit — agent-trust-hub — switchboard