Add Aomi to Security, Sandbox & Permissions#54
Conversation
The section's sandbox/permission entries gate code, shell, and network execution; none cover gating on-chain financial actions, a distinct harness trust-boundary problem. Aomi simulates each transaction before signing and signs non-custodially, so the pattern generalizes.
There was a problem hiding this comment.
Pull request overview
Note
Copilot was unable to run its full agentic suite in this review.
Adds a new “real repository worth studying” entry to the README’s agent/AI security/runtime resources list.
Changes:
- Added a new bullet linking to the Aomi Rust agent runtime repository with a brief rationale and stars badge.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
|
Hi @ai-boost, thanks for keeping this list sharp. I added Aomi under Security, Sandbox & Permissions. The reason I think it fits: every entry in that section gates code, shell, or network execution. None cover gating on-chain financial actions, which is a distinct trust-boundary problem. For an agent that transacts, the irreversible action is not Happy to tighten the wording, shorten the note, or move it to a different section if you think it lands better elsewhere. Thanks for taking a look. |
Adds Aomi to the Security, Sandbox & Permissions section.
Why it belongs: Every sandbox/permission entry in this section gates code, shell, or network execution (E2B, Daytona, OpenShell, forkd, the OWASP/NVIDIA references, etc.). None address gating on-chain financial actions, which is a distinct harness trust-boundary problem: for an agent that transacts, the irreversible action isn't
rm -rf, it's signing a transaction.Aomi is a Rust agent runtime where the permission boundary is the transaction: tools are compile-time-typed transactions, the runtime simulates each transaction's state effects before signing, and signing is non-custodial (the harness never holds keys). The generalizable pattern — typed action + pre-execution simulation + non-custodial authorization — applies to any chain, matching the list's vendor-agnostic-by-principle criterion.
One entry, one section, standard format.