The Agent Skills Directory

[SAFE]: The skill implements strong security guardrails (W009 and W011) that require explicit user approval for transactions and mandate treating all on-chain data as untrusted input. These instructions mitigate common AI agent risks such as unauthorized fund transfers and indirect prompt injection.
[SAFE]: Toolchain installation procedures for Agave and Surfpool use a 'download then execute' pattern but correctly advise the operator to inspect and verify the integrity of scripts before running them, rather than piping them directly to a shell.
[SAFE]: All identified external repository links and package dependencies point to the author's infrastructure (openlab-openos), trusted ecosystem organizations (Anza, Solana Foundation), or well-known development services. There are no signs of typosquatting or malicious obfuscation.

openverse-dev