The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents networking utilities (fetchJson, FetchClient) in fetch.md that ingest data from external APIs. This untrusted data represents a potential vector for indirect prompt injection if the agent processes the content without adequate sanitization or boundary markers.
Ingestion points: fetchJson and FetchClient in fetch.md ingest untrusted data from network endpoints.
Boundary markers: Absent from the documentation.
Capability inventory: Shell execution via pnpm (conventions.md), file and database writes via @ariestools/storage-adapters (packages.md), and network operations via fetch (fetch.md).
Sanitization: No explicit sanitization or escaping instructions for external API content are provided.
[EXTERNAL_DOWNLOADS]: The documentation recommends the installation of various packages from the @ariestools/* scope (e.g., @ariestools/sdk, @ariestools/express, @ariestools/threads) and third-party peers such as zod, undici, and @opentelemetry/api. These are legitimate dependencies for the Aries Tools development ecosystem.
[COMMAND_EXECUTION]: conventions.md provides shell commands for monorepo maintenance, such as pnpm xy build and pnpm sync-sdk-layout. These commands are intended for developers working within the sdk-js repository to synchronize monolith layouts and build packages.

ariestools-sdk