Posts tagged “security”
3 posts.
-
A command classifier should only run when nobody can be asked
opencode-guard adds permission modes, a static shell safety floor, credential guards and redaction to opencode, and asks a model only when no person can be.
-
Running coding agents unattended: an instruction is not a security boundary
A private control plane that syncs agent config across machines and runs scheduled agent jobs inside permission envelopes they cannot widen.
-
The worst agent incidents are the ones nobody narrates
An unattended agent reads my own coding-agent transcripts, decides which moments were real incidents, and files them in a ledger, exclusions included.