Creating your first ruleset
Stop repeating your standards to AI every session — write them once as a ruleset, attach it to your projects, and change them in one place.
If you find yourself telling your AI the same things at the start of every session — "use British spelling", "follow our coding standards", "don't commit secrets" — a ruleset is where those belong. You write them once, connect them to as many projects as you like, and every agent working on those projects follows them.
Create the ruleset
Go to Rulesets in the sidebar and choose New Ruleset.
Give it a name that says what it covers — "WordPress Development Standards", "Brand Voice", "API Conventions". Prefer several small, specific rulesets over one large one: they attach independently, so each project takes only what applies to it. One giant "Standards" ruleset forces every project to take everything.
A new ruleset affects nothing until you add rules and attach it.
Add rules
Choose Add rule. Each rule is one sentence carrying one instruction — if it needs "and" to join two obligations, make it two rules.
Pick the type deliberately:
| You want to say | Use |
|---|---|
| "always", "make sure you" | Must |
| "never", "under no circumstances" | Must not |
| "I'd rather", "unless there's a reason not to" | Prefer |
| "check our X before doing Y" | Consult |
Reach for Prefer more than you expect. If you can imagine a legitimate exception, it is a preference, not a requirement — and a workspace full of hard requirements that keep stopping the agent to ask trains people to ignore them.
Consult is the most useful and least obvious type. Instead of pasting your design system into a rule, keep it in a context library and add a consult rule pointing at it. The pointer is always present and costs almost nothing; the library only loads when the work calls for it.
Add a reason wherever one exists — "we removed Moment.js in v3.2 for bundle size". Rules with a stated reason get followed more reliably, and nobody deletes them six months later thinking they were arbitrary.
Is it actually a rule?
Rulesets hold instructions, not information. Read what you have written aloud:
- Telling someone to do something is a rule.
- Telling them how things are is context, and belongs in a context library.
"Must use our i18n functions" is a rule. "Billing runs through Stripe on three tiers" is a fact — put it in a library. If it is longer than a sentence, it is a document: put it in a library and write a consult rule pointing at it.
Attach it to a workflow
Open a workflow, go to the Rules tab, and choose Attach ruleset. From that point, every agent working in that workflow receives those rules.
Detaching is immediate and reversible. Attachment order does not matter.
See what your agents are actually following
From the same tab, choose See all rules in effect. This shows every rule that reaches your agents, grouped by which ruleset it came from, along with anything that is not applying and why — duplicates, rules replaced by a more specific one, archived rulesets, and rules held back by permissions.
It is the fastest way to answer "why did my agent do that?"
Ask your AI to write them for you
You do not have to author rules by hand. Tell your assistant "remember that we always use British spelling" and it can create the rule for you — choosing the right type, adding a reason, and putting it in a sensible ruleset. Every rule an AI writes is recorded as agent-authored in your activity feed and saved as a version, so you can always see what changed and undo it.
An AI cannot add to your workspace baseline, and cannot delete or rewrite existing rules. Those stay with you.