The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill demonstrates patterns for creating a certified key-value store that ingests untrusted data from external users, which could be used as a vector for indirect prompt injection if the stored data contains malicious instructions.
Ingestion points: The set(key, value), set_many(entries), and get(key) functions in the Rust and Motoko implementations accept user-supplied strings (SKILL.md).
Boundary markers: No explicit boundary markers or "ignore instructions" delimiters are used when processing or storing these strings.
Capability inventory: The skill enables writing to the canister's certified state via certified_data_set and serving that data back to users/agents in query calls.
Sanitization: The code snippets perform direct byte conversion of input strings without sanitizing for potentially malicious command or instruction sequences.
[EXTERNAL_DOWNLOADS]: The documentation specifies several external dependencies required for implementation.
Node.js packages: References @icp-sdk/core and @dfinity/certificate-verification as frontend requirements.
Rust crates: Includes ic-cdk, ic-certified-map, and ic-http-certification in the Cargo.toml examples.
Motoko packages: References the sha2 and ic-certification packages retrieved via the mops package manager.

certified-variables