Back to Oh My Openagent

Task 2 - atomic batch steering evidence

.omo/evidence/20260721-ulw-loop-gajae-adoption/task-2.md

4.19.12.1 KB
Original Source

Task 2 - atomic batch steering evidence

RED

Command: npm test -- steering-batch cli-steering-batch

Observed failure before implementation:

  • test/steering-batch.test.ts failed to import missing ../src/steering-batch.js.
  • parseSteeringProposals was not exported from cli-steering.ts.

GREEN / component gates

Commands:

  • Historical original gate: npm test -- steering-batch cli-steering-batch steering cli-steering, npm run typecheck, npm run build.
  • Reviewer-fix gate: npm test -- steering-batch cli-steering-batch validation-batch cli-validation-batch, npm run check, full npm test.

Observed:

  • Historical original gate: 6 test files passed, 78 tests passed; TypeScript strict check passed; build passed.
  • Reviewer-fix gate: 5 focused files / 22 tests passed; npm run check passed; full component suite passed 40 files / 418 tests.

Real-surface QA

Built CLI: dist/cli.js in fresh mktemp git repos.

Scenarios:

  1. steer --proposals-json with two valid proposals (annotate_ledger, revise_criterion).
    • Observed accepted batch result with results.length == 2.
    • Observed G001-goal-alpha criterion C001 scenario changed to observable revised scenario.
  2. steer --proposals-json with first valid add_subgoal and second invalid reorder_pending using missing.
    • Observed rejected batch result with unknown pending id.
    • Compared .omo/ulw-loop/goals.json before/after; byte string was unchanged, proving no partial plan mutation.
  3. Reviewer-fix QA re-ran rejected batch with two invalid items.
    • Observed goals byte-identical and exactly one steering_rejected ledger entry.
    • Observed message index 1: unknown pending id; index 2: split_subgoal requires target.
  4. Reviewer-fix QA re-ran accepted batch plus replay.
    • Observed accepted order G003,G001-goal-alpha,G002-goal-beta.
    • Observed replay all deduped with ledger count unchanged 3 -> 3, proving replay dedup produced no new mutation/audit.

Transcript summary printed by assertions:

  • Historical: batch accepted ok 2; batch rejection atomic ok unknown pending id.
  • Reviewer-fix: see reviewer-fix-transcript.txt.