One post per project: what problem it exists for, how it is built, and what remains unsolved.
What it actually takes to let an AI agent act on your behalf — and what two small, adversarially-reviewed libraries get you today. agent-guard is on PyPI as toolcall-authz; agent-warrant is the cross-org seal.
Two unglamorous pieces of infrastructure a solo operator running many agents actually needs — a workflow that doesn't skip steps, and a memory that doesn't quietly accumulate contradictions.
An OS-level security daemon for validators, signers, and bridge nodes: a one-second alert path for single hostile events, and a five-minute LLM pass that correlates sequences into attack patterns.
When two agents disagree, most systems pick a winner silently. Conflict Lens surfaces the disagreement as a first-class object you can inspect and resolve.