The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the arkiv-sync package and components of the Arkiv SDK hosted on unpkg.com and GitHub. These links are provided for version verification, protocol auditing, and implementation guidance, following standard developer practices for the platform.
[DYNAMIC_EXECUTION]: The OwnedMirror class in references/checkpoint-mirror.md implements a custom state serialization and restoration system (pack/unpack). This mechanism is used to persist the indexer's local state and rollback journal. The implementation is highly defensive, employing recursive type checks, regex-based validation for identifiers, and structural integrity checks to prevent unsafe deserialization of local data.
[INDIRECT_PROMPT_INJECTION]: The skill defines a data indexing surface that processes external blockchain logs.
Ingestion points: Data is ingested via client.getLogs and client.select within the readNativeBlock and readMarketplaceRowAt functions.
Boundary markers: The logic uses structured ABI decoding and strongly-typed SDK queries to establish clear boundaries between untrusted log data and the application projection.
Capability inventory: The skill manages an internal state projection and does not expose high-risk capabilities such as arbitrary command execution or network exfiltration based on ingested content.
Sanitization: The reference implementation performs rigorous schema and type validation on all incoming and restored data points to ensure they conform to the expected format.
[SAFE]: The skill demonstrates a strong commitment to security best practices, including canonical chain identity verification, atomic commitment of state updates, and clear warnings regarding the scope of application-side authorization. No malicious behavior or evasion techniques were detected.

arkiv-indexing