DocsStart
Goals
What an answer depends on: cost, depth, maker origin, the kind of data and the accounts held, the one question for each, and how an agent asks it.
Which model a feature should use, or what an agent should connect to, depends on what the app is for: what a call may cost, how hard the work is, where the data may go, which accounts the team already holds. These are its goals. Your coding agent plans the app with you; Rhumb answers the decisions it has dated facts for, and its answers are only as right as the goals they were asked under.
How an agent asks
The skill and the MCP server give an agent these rules. An agent that follows them asks you little: most goals are already in your plan or your repository.
- Ask only the goals the next call depends on, one at a time, and at most five in a planning session; state the rest as assumptions when showing the answer.
- Look first: the plan or spec already written, then the repository (dependency, environment variable and config file names, never values), then the conversation. Ask the user only what none of these answers.
- Use the harness's structured question tool when it has one; otherwise ask a numbered question in chat, with the default first.
- A project-wide answer for cost and depth stands until the user overrides it for a feature.
- Never invent a goal the user did not state, and never send what they said in their own words: goals travel only as the fixed values each goal lists.
The goals
Each goal has one question and the parameters its answer is sent as. A goal that sends nothing yet is one the API does not take yet; it is listed so an agent can say it was not considered.
cost
How much a call, or a user, may cost.
Sent as optimize, max_price, budget, calls_per_user_month on resolve, plan, stack. On stack, the app's goals go by their own names at the top (data_kind, users_in, scale, clients), a model's parameters in model.constraints and providers in model.providers; infrastructure vendors are infra.vendors.
What should this feature optimize for?
- Value: the best for its price (the default offered) →
optimize=value - Lowest cost that is still good →
optimize=cost - Best quality, whatever it costs →
optimize=quality - A cap per call, stated →
max_price=<usd>
Do you know roughly how often one user will use this in a month?
- Not yet: price one call (the default offered)
- A number, stated →
calls_per_user_month=<calls>
depth
How much quality or reasoning a feature needs.
Sent as optimize, quality_margin, min_quality, budget, shape, needs on resolve, plan.
How hard is the work this feature does?
- Routine: a fast model does it (the default offered) →
budget=cheap - Mixed: try a cheap model first and escalate →
shape=escalation - Hard: needs a model that reasons →
needs=reasoning
maker_origin
Where a model's maker may be based.
Sent as origin, origin_strict on resolve, plan, stack. On stack, the app's goals go by their own names at the top (data_kind, users_in, scale, clients), a model's parameters in model.constraints and providers in model.providers; infrastructure vendors are infra.vendors.
Any limit on where a model's maker is based?
- No limit (the default offered)
- EU only →
origin=eu - US only →
origin=us - Not China →
origin=not:cn
users_in
Where the app's users are.
Sent as users_in on stack, platform. On stack, the app's goals go by their own names at the top (data_kind, users_in, scale, clients), a model's parameters in model.constraints and providers in model.providers; infrastructure vendors are infra.vendors. Recorded with the answer and listed as assumed when left out; no rule reads it yet, so it never changes an answer and is never asked.
Where will your users be?
- Worldwide (the default offered) →
users_in=worldwide - EU included →
users_in=eu - US only →
users_in=us - Other, stated by region
data_region
Where data may be processed and stored.
Sent as regions on resolve, plan, platform. A region filters only when at least half the offerings on the providers named state where they are served; below that the answer warns region_not_filtered and removes nothing.
Must data stay in a region?
- No (the default offered)
- EU →
regions=eu - US →
regions=us - Other, stated →
regions=<region>
languages
The languages users write in.
Sent as languages on resolve, plan. A language other than English puts the non-English quality signal first for a purpose that lists it; an answer not compared on it warns language_signal_unavailable.
Which languages will users write in?
- English only (the default offered) →
languages=en - Several, stated →
languages=<codes>
data_kind
Whether the app handles health, financial or children's data, or serves a covered entity.
Sent as needs, data_kind on resolve, plan, stack, platform. On stack, the app's goals go by their own names at the top (data_kind, users_in, scale, clients), a model's parameters in model.constraints and providers in model.providers; infrastructure vendors are infra.vendors.
What kind of personal data does it handle?
- None sensitive (the default offered)
- Health or fitness →
needs=zdr,no_training - Financial →
needs=zdr,no_training - For a healthcare provider or insurer (HIPAA) →
needs=zdr,no_training
scale
Prototype, launch, or a stated number of users.
Sent as scale on stack, platform. On stack, the app's goals go by their own names at the top (data_kind, users_in, scale, clients), a model's parameters in model.constraints and providers in model.providers; infrastructure vendors are infra.vendors.
What stage is this?
- Prototype for yourself (the default offered) →
scale=prototype - Launching to the public →
scale=launch - Already live, with a user count stated →
scale=live
accounts
The model providers, infrastructure vendors and platforms already held.
Sent as providers, infra.vendors, accounts on resolve, plan, stack, platform. On stack, the app's goals go by their own names at the top (data_kind, users_in, scale, clients), a model's parameters in model.constraints and providers in model.providers; infrastructure vendors are infra.vendors.
Which of these do you already have an account with? (found in the repository, shown for correction)
- These, as found (the default offered)
- These, corrected
clients
Where it runs for the user: iOS, Android, web, desktop or server only.
Sent as clients on stack, platform. On stack, the app's goals go by their own names at the top (data_kind, users_in, scale, clients), a model's parameters in model.constraints and providers in model.providers; infrastructure vendors are infra.vendors.
Where does it run for the user?
- Web (the default offered) →
clients=web - iOS →
clients=ios - Android →
clients=android - Several →
clients=<list>
speed
Rhumb keeps no latency data, so no answer says one model is faster than another, and an agent following the skill does not ask how fast a feature must be. When speed matters, measure it on your own calls.