Agent tooling
Throwaway Jev demo: route coding tasks to Grok Build or Codex Astra
import { experimental_evaluate as evaluate } from 'ai';The link points at the commit we read, so the line number still holds.
These question sets are lifted from this project's source exactly as written.
from router.mjs:82
complexityscoreRate the reasoning and implementation complexity of this task.
0 — mechanical: one obvious localized change with direct verification1 — contained: limited reasoning across a few known files or steps2 — systemic: multiple components, unclear diagnosis, or meaningful tradeoffs3 — high consequence: architecture, security, migration, or broad ambiguous workneedsPlanningtrue/falseShould a capable engineer plan or investigate before changing code?
import { experimental_evaluate as evaluate } from 'ai';
const { answers } = await evaluate({
model: 'typesafe-ai/jev',
state,
questions: {
complexity: { type: 'score', instructions: 'Rate the reasoning and implementation complexity of this task.',
criteria: ['mechanical: one obvious localized change with direct verification', 'contained: limited reasoning across a few known files or steps', 'systemic: multiple components, unclear diagnosis, or meaningful tradeoffs', 'high consequence: architecture, security, migration, or broad ambiguous work'] },
needsPlanning: { type: 'boolean', instructions: 'Should a capable engineer plan or investigate before changing code?' },
},
});