ms-ai-architect/tests/kb-eval/test-score-skill.test.mjs
Kjell Tore Guttormsen c1a09062d4 feat(ms-ai-architect): Spor D steg 2 — score-skill CLI (--json/--gate/--skill) (TDD)
Byggekloss for både SessionStart-surfacing og lifecycle-gate (D3).

- eval.mjs: ekstraher buildReport() (eksportert, testbar) ut av main() —
  samler hele korpus-rapporten (deterministisk + K10 + merget judge-cache).
- lib/skill-score.mjs: formatScoreReport() (ren) — sortert forbedringsrapport
  med score, status, floored/provisional/ujudget-tags og ⚑gulv-markering.
- score-skill.mjs (NY CLI): buildReport → scoreReport → render/json/gate.
  --skill scorer hele korpuset (K10 trenger alle søsken) men rapporterer/gater
  kun den valgte → inkrementell enkelt-skill-scoring for D3 lifecycle-gate.
- docs/development.md: ny «Skill-kvalitetsscore (Spor D)»-seksjon.

Live baseline: advisor 91, governance/security 96 (OK); engineering +
infrastructure 89 (floored av K10-gulv) UNDER mål. Gate exit=1.

Tester: kb-eval 134→142 (+8). validate 239, kb-update 316 uendret grønt.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-23 17:16:44 +02:00

113 lines
5.2 KiB
JavaScript

// tests/kb-eval/test-score-skill.test.mjs
// Tests for the score-skill CLI building blocks (Spor D, step 2):
// - eval.mjs buildReport() assembles the full corpus report (deterministic +
// K10 sibling-overlap + merged judge cache).
// - lib/skill-score.mjs formatScoreReport() renders the sorted improvement
// report (pure).
// - the score-skill.mjs CLI: --json emits the scoreReport, --gate sets a
// non-zero exit when any skill is below target, --skill scopes the output.
import { test } from 'node:test';
import assert from 'node:assert/strict';
import { execFileSync } from 'node:child_process';
import { fileURLToPath } from 'node:url';
import { dirname, join } from 'node:path';
import { buildReport } from '../../scripts/kb-eval/eval.mjs';
import { scoreReport, formatScoreReport } from '../../scripts/kb-eval/lib/skill-score.mjs';
const __dirname = dirname(fileURLToPath(import.meta.url));
const CLI = join(__dirname, '..', '..', 'scripts', 'kb-eval', 'score-skill.mjs');
test('buildReport — assembles the full corpus with N1-N5 on every skill', () => {
const report = buildReport();
assert.ok(Array.isArray(report.skills) && report.skills.length === 5, `got ${report.skills?.length} skills`);
for (const s of report.skills) {
const d = s.deterministic;
assert.ok(d.N1_nameValidity && d.N4_refToc, `${s.name} missing N1/N4`);
assert.ok(d.K10_siblingScopeOverlap, `${s.name} missing K10 (sibling overlap not attached)`);
}
});
test('scoreReport — over the real corpus returns a score per skill and a below-target set', () => {
const result = scoreReport(buildReport());
assert.equal(result.target, 90);
assert.equal(result.scored.length, 5);
for (const s of result.scored) {
assert.ok(Number.isFinite(s.score), `${s.name} score not finite`);
assert.equal(typeof s.meetsTarget, 'boolean');
}
// below = subset that does not meet the target
assert.ok(result.below.every((s) => !s.meetsTarget));
});
test('scoreReport — a synthetic below-target skill lands in below[]', () => {
const synthetic = {
skills: [
// one perfect-ish, one with a failing K10 floor (capped below 90)
{
name: 'good', deterministic: {
K2_descriptionFormat: { pass: true }, K3_bodyLength: { bodyLines: 100, pass: true },
K5_progressiveDisclosure: { namedRatio: 0.5, pass: true }, K6_routingTable: { pass: true },
refCountConsistency: { consistent: true }, K10_siblingScopeOverlap: { maxCombined: 1, pass: true, threshold: 7 },
N1_nameValidity: { pass: true }, N2_descriptionLength: { pass: true },
N3_refNestingDepth: { pass: true }, N4_refToc: { ratio: 1, pass: true, withToc: 1, largeFiles: 1 },
N5_forwardSlashPaths: { pass: true },
}, judge: null,
},
{
name: 'bad', deterministic: {
K2_descriptionFormat: { pass: true }, K3_bodyLength: { bodyLines: 100, pass: true },
K5_progressiveDisclosure: { namedRatio: 0.5, pass: true }, K6_routingTable: { pass: true },
refCountConsistency: { consistent: true }, K10_siblingScopeOverlap: { maxCombined: 9, pass: false, threshold: 7, worstSibling: 'good' },
N1_nameValidity: { pass: true }, N2_descriptionLength: { pass: true },
N3_refNestingDepth: { pass: true }, N4_refToc: { ratio: 1, pass: true, withToc: 1, largeFiles: 1 },
N5_forwardSlashPaths: { pass: true },
}, judge: null,
},
],
};
const result = scoreReport(synthetic);
assert.equal(result.below.length, 1);
assert.equal(result.below[0].name, 'bad');
});
test('formatScoreReport — renders skill names, the target, and improvement fixes', () => {
const result = scoreReport(buildReport());
const text = formatScoreReport(result);
assert.match(text, /ms-ai-advisor/);
assert.match(text, /90/); // the target appears
// at least one improvement fix line is rendered for a below-target skill
if (result.below.length > 0) {
assert.ok(/↳|→/.test(text), 'expected improvement bullets in the report');
}
});
test('CLI --json — emits a parseable scoreReport with five scored skills (exit 0)', () => {
const out = execFileSync('node', [CLI, '--json'], { encoding: 'utf8' });
const parsed = JSON.parse(out);
assert.equal(parsed.scored.length, 5);
assert.equal(parsed.target, 90);
});
test('CLI --skill — scopes the output to a single skill', () => {
const out = execFileSync('node', [CLI, '--skill', 'ms-ai-advisor', '--json'], { encoding: 'utf8' });
const parsed = JSON.parse(out);
assert.equal(parsed.scored.length, 1);
assert.equal(parsed.scored[0].name, 'ms-ai-advisor');
});
test('CLI --gate — non-zero exit path: an unreachable target fails the gate (corpus-independent)', () => {
let code = 0;
try {
execFileSync('node', [CLI, '--gate', '101'], { encoding: 'utf8', stdio: 'pipe' });
} catch (e) {
code = e.status;
}
assert.equal(code, 1, 'no skill can score 101 -> gate must fail');
});
test('CLI --gate — zero exit path: a target of 0 passes the gate (corpus-independent)', () => {
// exits 0; execFileSync throws only on non-zero, so reaching here is the pass
execFileSync('node', [CLI, '--gate', '0'], { encoding: 'utf8', stdio: 'pipe' });
assert.ok(true);
});