Writing Effective AOPs
The highest-leverage thing you control when building an Agent.
Choosing How Your Agent Starts
Manual Starts, schedules, triggers, and Queues each fit a different shape of work.
Designing Work Around Queues and Cases
Producer/consumer design so every Case ends somewhere definite.
Connecting Agents with Queues
Split a growing process into stages connected by Queues — every item gets its own Run, the thousandth processed as reliably as the first.
Where Knowledge Belongs
AOP, Skills, Files, or Memory: four places to teach an Agent something, each with different scope.
Iterating on a Live Agent
Editing the active revision changes production immediately. Fork a draft, test on real work, make it active deliberately.
Let a Skill Do the Math
When results must be exact and auditable, package the calculation as a script inside a custom Skill.
Wrap a REST API in a Skill
No native Connection, but the system has an HTTP API? Package the call as a script inside a custom Skill.
Browsing That Survives UI Changes
Depend on what things are, not where they are — stable anchors, recorded Skills, and verify-after-write habits.
Connections and Logins That Don't Break
Set up access that survives teammates, departures, and expired credentials.
One Builder, Many Users
Why an Agent works for its builder and fails for everyone else — the two access gates and the sharing model.
When the Agent Should Ask, Not Act
Approval gates fire only where the AOP asks. Which actions earn one, whose inbox it lands in, and designing for the wait.
Make Every Run Auditable
Automatic Evaluations, durable outputs, and the alerting surface — so “what did the Agent do?” is a link, not an investigation.
Tips and Patterns
A cross-referenced index of practical guidance from across the docs.
We’re actively expanding this section. Check back for more detailed guidance.