Advisors

Advisors are repo-specific agents you register so Cadence can ask them for a second opinion during the thinking stages. Think of them as named experts on your codebase (a “database advisor,” a “security advisor”) that you can pull into a brainstorm or design conversation when a question is genuinely architectural.

They are entirely opt-in, off by default, and they never write into your documents.

Registering advisors

Advisors live in your repo’s .cadence/config.yaml under the advisors: key. There are two settings:

advisors:
  enabled: false   # master switch: true to allow consultation at all
  roster: []       # the named agents Cadence may consult

• advisors.enabled: the master switch. While it’s false (the default), Cadence never offers to consult anyone.
• advisors.roster: the list of agents available to consult. Add the names of the repo-specific agents you want on call.

To turn the feature on, set enabled: true and add one or more names to roster. That’s all the configuration there is.

Off until you opt in

Out of the box advisors.enabled is false and roster is empty, so a fresh repo never sees an advisor prompt. You decide if and when advisors become part of your workflow.

When advisors are consulted

Advisors come up in exactly two stages (/c-brainstorm and /c-design) and only at the right moment.

During the Q&A, when a question is genuinely architectural (“does this approach scale?”, “is this the right boundary?”), Cadence offers to consult your advisors. It’s an explicit, case-by-case offer, something like:

Want me to consult database-advisor, security-advisor? Token-heavy.

You opt in (or not) for that one question. The offer is flagged as token-heavy on purpose: consulting advisors dispatches additional agents and costs more, so Cadence makes the cost visible and lets you decide it’s worth it for this particular decision.

Not for every question

Advisors are reserved for real architectural forks, not routine choices. Cadence won’t offer them on small decisions, and you shouldn’t reach for them there either: the value (and the cost) is in the hard calls.

Their opinions feed the Q&A, never the artifact

This is the load-bearing rule. When you opt in, Cadence dispatches the advisors (in parallel) and their opinions flow back into the question-and-answer conversation. They inform what gets asked next and what trade-offs surface.

They do not get written into the stub or design directly. If two advisors disagree, Cadence surfaces both views. It doesn’t pick a winner. You adjudicate. The advisor is a consultant in the room, not an author with edit rights.

That keeps the artifact yours: every decision recorded in a design’s decisions log is one you made, possibly informed by an advisor, but never authored by one.

Reference

• Full config schema and defaults: Configuration reference.
• The stages that consult advisors: /c-brainstorm and /c-design.