@manya/reflection

Evaluates outputs, critiques plans, and turns failures into lessons that force a revised plan — not just a logged note nobody reads.

Composed from @manya/memory (where lessons live as semantic facts) and @manya/cortex (asked to re-plan once a lesson has been learned).

Usage

import { memory } from '@manya/memory';
import { createCortex } from '@manya/cortex';
import { reflection } from '@manya/reflection';

const store = memory.createMemoryStore('ara');
const cortex = createCortex(store);
const refl = reflection.createReflection(store, cortex);

reflection.evaluateAgainstIntent(refl, 'schedule a dentist appointment', 'Booked for Tuesday.');
// { passes: true, ... }

const { lesson, revisedPlan } = reflection.learnFromFailure(
  refl,
  'derive a key',
  'no passphrase was provided'
);
// lesson is now in semantic memory, and revisedPlan is a fresh cortex.plan()
// call made with that lesson already recorded.

Why this composition

The point of "learning from failure" is that the next plan is actually different. Reflection doesn't just write the lesson down — it immediately calls back into Cortex to produce a new plan, so the lesson has a chance to change the confidence score and steps of what happens next.