The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides an interface for storing and retrieving user-controlled data, which represents a potential injection surface if the data is processed by an AI agent or application without proper sanitization.
Ingestion points: User-supplied values and keys are ingested via frontend methods like setValue and backend endpoints such as insert_encrypted_value_with_metadata.
Boundary markers: The implementation relies on the ic-vetkeys library's internal access control and client-side encryption. No explicit input validation or boundary markers for the data content are demonstrated in the snippets.
Capability inventory: The skill facilitates writing to canister stable memory and performing network operations via the Internet Computer's HttpAgent.
Sanitization: There is no mention of sanitization or escaping of the stored data in the provided documentation.
[PERSISTENCE]: The skill utilizes browser storage for cryptographic material to enable cross-session persistence, which is a standard requirement for the described use case.
Evidence: The frontend implementation uses IndexedDbDerivedKeyMaterialCache to store derived key material in IndexedDB.
Mitigation: The skill includes clear instructions to call clearCache() during logout or identity switches to ensure sensitive key material does not persist beyond the active session.
[EXTERNAL_DOWNLOADS]: The skill references several external packages for frontend and backend development within the vendor's ecosystem.
Evidence: Dependencies include @icp-sdk/vetkeys, @icp-sdk/core, and the Rust/Motoko library ic-vetkeys.
Context: The skill provides guidance on migrating from legacy @dfinity/ namespaces to current @icp-sdk/ packages, consistent with the vendor's updated SDK structure.

encrypted-maps