The Agent Skills Directory

[SAFE]: The skill defines a comprehensive security and privacy policy for the Arkiv ecosystem. It focuses on teaching best practices such as off-chain secret management and verifiable publication trust boundaries.
[INDIRECT_PROMPT_INJECTION]: The skill identifies and provides mitigations for potential indirect prompt injection surfaces when processing external entity data.
Ingestion points: The trustedReadonlyPost and readPublications functions in references/trust-boundary.md process entity payloads from the Arkiv network.
Boundary markers: The publicationMessages function implements a defense-in-depth approach by wrapping ingested data in a system-role message that explicitly instructs the agent to ignore any instructions embedded within the record.
Capability inventory: The provided reference code is limited to cryptographic indexing (HMAC), schema validation, and safe DOM manipulation.
Sanitization: The skill recommends strict field-level validation and mandates using textContent or framework-level escaping to prevent cross-site scripting (XSS) or model manipulation.
[EXTERNAL_DOWNLOADS]: The skill references source code and consumer guides for the @arkiv-network/sdk and arkiv-encryption library from unpkg.com and github.com. These are legitimate vendor resources used for providing technical documentation and implementation guidance.

arkiv-security-trust