#guardrails
Every post tagged "guardrails".
39 posts · showing 31–39
-
Designing a Kill Switch That Works Mid-Run
Killing the process leaves side effects in an unknown state. A stop that works checks between steps — before every model call and before every tool call — and the difference matters most at 2am.
-
The Escalation Path Is Part of the Design
An agent with no defined way to stop will improvise one, and its improvisation is a confident answer. Escalation deserves a tool, a destination, and a metric — and one field that stops the human starting over.
-
The On-Call Agent: What It Should and Shouldn't Touch During an Incident
An incident is when your system is least understood — a poor moment for an autonomous actor, an excellent one for an autonomous investigator. The line between them is read access and write access.
-
Indirect Prompt Injection: When the Tool Result Is the Attack
The attack doesn't come from your user — it comes from the page, ticket, or document your agent fetched. Three conditions have to hold for it to work, and breaking any one is an architectural choice, not a prompt.
-
Where to Put the Approval: Human-in-the-Loop Is a Design, Not a Checkbox
A confirm dialog nobody reads has all the latency of oversight and none of the protection. Four placements, a routing rule based on reversibility, and the metric that tells you your gate has become a rubber stamp.
-
Prompt, Tool, or Code? Where Each Piece of Logic Belongs
A rule the agent keeps breaking usually isn't a prompting failure — it's deterministic logic filed under "advice." Two questions place any behavior in the right one of three homes.
-
Agent Adoption Stalls at Trust, Not Capability
An agent whose errors cluster somewhere describable beats a more accurate one whose errors are scattered. Ask your users when they double-check its work — the answer is your real roadmap.
-
The Read-Only Agent That Wasn't: Auditing What Your Tools Can Actually Reach
A tool named `check_service_health` woke someone at 3am. Read-only is a claim about effects at the far end of every call, and there's a short audit that either proves it or shows you what you actually deployed.
-
The Confused Deputy in Your Agent: When the Tool Trusts the Model Too Much
Your refund tool checks whether the agent is allowed to issue refunds. Nothing checks whether the customer asking for one is entitled to it — and that gap opens without any jailbreak at all.