The Agent Skills Directory

Prompt Injection (HIGH): The skill guides the creation of tools that ingest untrusted user parameters and network data which are then processed by the AI agent. Ingestion points: createTokenParameters and getTokenInfoParameters defined in the examples/token-plugin/tools/ directory. Boundary markers: Absent in the example tool description prompts. Capability inventory: handleTransaction (blockchain mutations) in create-token.ts and fetch (network access) in get-token-info.ts. Sanitization: Runtime validation via Zod schemas is present for type checking, but lacks semantic filtering to prevent instruction injection within valid string fields.
Data Exposure & Exfiltration (LOW): Example tool get-token-info.ts performs network requests to a non-whitelisted domain (mirrornode.hedera.com). Evidence: fetch call used to query the Hedera Mirror Node API. Severity: LOW as it targets a standard infrastructure domain for the skill's purpose and does not access sensitive local files.

Hedera Plugin Creation