Udit Raj 75616bf73c Add evals harness, CI, contributing guide, badges, topics ...

- bench/run-evals.ts — ADHD vs single-shot baseline across problem set,
  LLM-as-judge scoring on breadth/novelty/trap-detection/actionability/
  builder-usefulness with A/B order randomized per problem
- bench/problems.json — 6 open-ended engineering problems
- bench/judge.ts — separate critic pass with skeptical-staff-eng prompt
- EVALS.md — populated: ADHD 5W / 1L / 0T vs baseline
    breadth +4.17, novelty +5.17, trap_detection +7.67,
    actionability +3.00, builder_usefulness +0.83 (mean Δ across 6 problems)
- .github/workflows/ci.yml — typecheck + build on Node 18/20/22
- .github/workflows/evals.yml — manual-trigger eval run, artifacts uploaded,
  EVALS.md committed back to main
- tsconfig.bench.json — typecheck covers bench/ too
- CONTRIBUTING.md — frame-authoring guide, eval-problem guide, loop invariants
- README badges (CI, Evals, npm, license, Node), Evals section

2026-05-25 16:59:31 +05:30

..

baseline.ts

Initial commit — ADHD, a skill for coding agents

2026-05-25 16:31:32 +05:30

judge.ts

Add evals harness, CI, contributing guide, badges, topics

2026-05-25 16:59:31 +05:30

problems.json

Add evals harness, CI, contributing guide, badges, topics

2026-05-25 16:59:31 +05:30

results.json

Add evals harness, CI, contributing guide, badges, topics

2026-05-25 16:59:31 +05:30

run-evals.ts

Add evals harness, CI, contributing guide, badges, topics

2026-05-25 16:59:31 +05:30