Prototypes & Scenarios

This week in rngo 0.40.0, I added prototypes, which enable parameter sharing among effects, and scenarios, which make it easy to eval different variants of a single spec.

#Prototypes

Often, multiple effects end up sharing the same channel, start and end. A prototype lets you deduplicate these:

prototypes:
  backfill:
    channel: db
    start: now - years(2)
    end: now
effects:
  user.create:
    prototype: backfill
    trigger: hz(5, hour)
    generator: ...

Here the user.create effect now uses the backfill prototype's channel, start and end.

Effects that do not specify a named prototype will get a default one, which takes start and end from the top-level spec.

#Scenarios

But more importantly than sharing parameters, prototypes enable scenarios, which override parameters upon request:

prototypes:
  phaseOne:
    start: now - years(2)
    end: now - years(1)
    scenarios:
      small:
        start: now - months(6)
        end: now - months(3)
  phaseTwo:
    start: now - years(1)
    end: now
    scenarios:
      small:
        start: now - months(3)

The scenario can be selected during evaled with the --scenario flag:

❯ rngo eval --scenario small

Here, the time bounds from the small scenarios will by used for every effect that references either the phaseOne or phaseTwo prototypes. So a simple flag produces a smaller, but logically consistent variant of the spec.

See the prototype reference and the rngo eval reference for more details.

#Agent Skill

rngo agent 0.6.0 updates the rngo skill's references so your coding agent knows about prototypes and scenarios.

You can update your skills by running:

❯ rngo skills install