brain/tools/safe-baseline-metering.test.mjs

// tools/safe-baseline-metering.test.mjs
import { describe, it, expect } from 'vitest';
import {
  RESET_MARKERS, isResetMarker, DEFAULT_THRESHOLDS, deriveTaskId,
  keywordOverlapCount, shouldInheritTaskId, newCounterState,
  incrementCounter, evaluateThresholds, isMutatingForBaseline,
} from './safe-baseline-metering.mjs';

describe('isResetMarker', () => {
  it('detects "новая задача" case-insensitive', () => {
    expect(isResetMarker('Окей, НОВАЯ ЗАДАЧА теперь')).toBe(true);
  });
  it('returns false for unrelated prompt', () => {
    expect(isResetMarker('почини баг в парсере')).toBe(false);
  });
});

describe('deriveTaskId', () => {
  it('is deterministic 16-hex for same input', () => {
    const a = deriveTaskId('fix foo');
    const b = deriveTaskId('fix foo');
    expect(a).toBe(b);
    expect(a).toMatch(/^[0-9a-f]{16}$/);
  });
  it('differs for different input', () => {
    expect(deriveTaskId('a')).not.toBe(deriveTaskId('b'));
  });
});

describe('keywordOverlapCount / shouldInheritTaskId', () => {
  it('counts intersection', () => {
    expect(keywordOverlapCount(['foo', 'bar', 'baz'], ['foo', 'bar', 'qux'])).toBe(2);
  });
  it('inherits when overlap>=2 and no reset marker', () => {
    expect(shouldInheritTaskId(['foo', 'bar'], ['foo', 'bar'], 'edge case')).toBe(true);
  });
  it('does NOT inherit on reset marker', () => {
    expect(shouldInheritTaskId(['foo', 'bar'], ['foo', 'bar'], 'новая задача')).toBe(false);
  });
  it('does NOT inherit when overlap<2', () => {
    expect(shouldInheritTaskId(['foo'], ['foo', 'x'], 'continue')).toBe(false);
  });
});

describe('incrementCounter', () => {
  const start = newCounterState({ taskId: 't', startedAtIso: '2026-05-29T00:00:00Z', firstPromptExcerpt: 'p' });

  it('increments Read counter', () => {
    expect(incrementCounter(start, 'Read').counts.Read).toBe(1);
  });
  it('maps TodoWrite to TodoWrite_writes', () => {
    expect(incrementCounter(start, 'TodoWrite').counts.TodoWrite_writes).toBe(1);
  });
  it('returns state unchanged for non-metered tool (Edit)', () => {
    const result = incrementCounter(start, 'Edit');
    expect(result.counts.Read).toBe(0);
  });
  it('is immutable — original state not mutated', () => {
    incrementCounter(start, 'Read');
    expect(start.counts.Read).toBe(0);
  });
});

describe('evaluateThresholds — warn/soft_flag', () => {
  function stateWith(patches) {
    const s = newCounterState({ taskId: 't', startedAtIso: '2026-05-29T00:00:00Z', firstPromptExcerpt: 'p' });
    return { ...s, counts: { ...s.counts, ...patches } };
  }

  it('Read=30 → soft_flag', () => {
    const s = stateWith({ Read: 30 });
    expect(evaluateThresholds(s, 'Read', false).action).toBe('soft_flag');
  });
  it('Read=29 → allow', () => {
    const s = stateWith({ Read: 29 });
    expect(evaluateThresholds(s, 'Read', false).action).toBe('allow');
  });
  it('Grep=15 → soft_flag', () => {
    const s = stateWith({ Grep: 15 });
    expect(evaluateThresholds(s, 'Grep', false).action).toBe('soft_flag');
  });
});

describe('evaluateThresholds — hard threshold, safe-baseline tool stays allowed', () => {
  function stateWith(patches) {
    const s = newCounterState({ taskId: 't', startedAtIso: '2026-05-29T00:00:00Z', firstPromptExcerpt: 'p' });
    return { ...s, counts: { ...s.counts, ...patches } };
  }

  it('Read=60 → allow (continuation reading)', () => {
    const s = stateWith({ Read: 60 });
    expect(evaluateThresholds(s, 'Read', false).action).toBe('allow');
  });
  it('Glob=20 → allow', () => {
    const s = stateWith({ Glob: 20 });
    expect(evaluateThresholds(s, 'Glob', false).action).toBe('allow');
  });
});

describe('evaluateThresholds — mutating hard-block', () => {
  function stateWith(patches) {
    const s = newCounterState({ taskId: 't', startedAtIso: '2026-05-29T00:00:00Z', firstPromptExcerpt: 'p' });
    return { ...s, counts: { ...s.counts, ...patches } };
  }

  it('Read=60, Edit → hard_block with reason containing Read=60', () => {
    const s = stateWith({ Read: 60 });
    const r = evaluateThresholds(s, 'Edit', false);
    expect(r.action).toBe('hard_block');
    expect(r.reason).toContain('Read=60');
  });
  it('Read=60, Bash → hard_block', () => {
    const s = stateWith({ Read: 60 });
    expect(evaluateThresholds(s, 'Bash', false).action).toBe('hard_block');
  });
  it('Read=59, Edit → allow (no counter at hard threshold)', () => {
    const s = stateWith({ Read: 59 });
    expect(evaluateThresholds(s, 'Edit', false).action).toBe('allow');
  });
  it('Read=60, Edit, skill=true → allow (skill match overrides)', () => {
    const s = stateWith({ Read: 60 });
    expect(evaluateThresholds(s, 'Edit', true).action).toBe('allow');
  });
  it('TodoWrite_writes=15, Write → hard_block with reason containing TodoWrite_writes=15', () => {
    const s = stateWith({ TodoWrite_writes: 15 });
    const r = evaluateThresholds(s, 'Write', false);
    expect(r.action).toBe('hard_block');
    expect(r.reason).toContain('TodoWrite_writes=15');
  });
});

describe('evaluateThresholds — skillMatched short-circuit', () => {
  it('Read=100, skill=true → allow regardless', () => {
    const s = newCounterState({ taskId: 't', startedAtIso: '2026-05-29T00:00:00Z', firstPromptExcerpt: 'p' });
    const big = { ...s, counts: { ...s.counts, Read: 100 } };
    expect(evaluateThresholds(big, 'Read', true).action).toBe('allow');
  });
});

describe('evaluateThresholds — LS and AskUserQuestion warn thresholds', () => {
  function stateWith(patches) {
    const s = newCounterState({ taskId: 't', startedAtIso: '2026-05-29T00:00:00Z', firstPromptExcerpt: 'p' });
    return { ...s, counts: { ...s.counts, ...patches } };
  }

  it('LS=10, skill=false → soft_flag', () => {
    const s = stateWith({ LS: 10 });
    expect(evaluateThresholds(s, 'LS', false).action).toBe('soft_flag');
  });
  it('AskUserQuestion=2, skill=false → soft_flag', () => {
    const s = stateWith({ AskUserQuestion: 2 });
    expect(evaluateThresholds(s, 'AskUserQuestion', false).action).toBe('soft_flag');
  });
});

describe('keywordOverlapCount — deduplication', () => {
  it('dedupes duplicates in a: ["router","router","gate"] ∩ ["router","gate"] = 2', () => {
    expect(keywordOverlapCount(['router', 'router', 'gate'], ['router', 'gate'])).toBe(2);
  });
});

describe('DEFAULT_THRESHOLDS — deep freeze', () => {
  it('outer object is frozen', () => {
    expect(Object.isFrozen(DEFAULT_THRESHOLDS)).toBe(true);
  });
  it('nested Read threshold object is frozen', () => {
    expect(Object.isFrozen(DEFAULT_THRESHOLDS.Read)).toBe(true);
  });
  it('nested AskUserQuestion threshold object is frozen', () => {
    expect(Object.isFrozen(DEFAULT_THRESHOLDS.AskUserQuestion)).toBe(true);
  });
});