d365fo-mcp 1.0.0

package/dist/eval/improver/caseMining.js

@@ -0,0 +1,64 @@
+/**
+ * Case mining (docs/AGENT_EVAL_LOOP.md §9, "Self-improving upgrades"). VM-free.
+ *
+ * "The catalog grows itself" — a real tool-defect/friction point hit during
+ * ANY agent session (not just a controlled eval run) is a candidate eval case:
+ * if it bit a real session once, it's worth a permanent regression case so a
+ * future fix (or future regression) is caught by the loop automatically. This
+ * module is the converter: given a short structured description of what went
+ * wrong, produce a schema-conformant `eval/cases/*.json` DRAFT.
+ *
+ * The draft is deliberately a skeleton, not a finished case: it carries
+ * `golden_pending: true` (no golden captured yet — that needs a VM run per
+ * §6.4) and enters the `holdout` split (§10, new cases are held out by
+ * default). A human/improver-agent still needs to refine the `instruction`
+ * into something precise enough to reproduce deterministically, and a VM run
+ * captures the golden before the case is "real". Mining only saves the
+ * bookkeeping (id, schema shape, file placement) so a real failure doesn't
+ * just get fixed and forgotten.
+ */
+import * as fs from 'fs';
+import * as path from 'path';
+/** Lowercase, alphanumeric-and-dashes slug matching the case id pattern `^[a-z0-9-]+$`. */
+export function slugify(text) {
+    return text
+        .toLowerCase()
+        .replace(/[^a-z0-9]+/g, '-')
+        .replace(/^-+|-+$/g, '')
+        .replace(/-{2,}/g, '-')
+        .slice(0, 60);
+}
+/** Build a draft case spec from a mined failure description. Does not touch disk. */
+export function mineCaseFromFailure(input) {
+    const slug = input.idSlug ? slugify(input.idSlug) : slugify(input.title);
+    if (slug.length < 3) {
+        throw new Error(`Could not derive a usable case id slug from "${input.title}" — pass idSlug explicitly.`);
+    }
+    const id = `L${input.tier}-${slug}`;
+    return {
+        id,
+        title: input.title,
+        tier: input.tier,
+        instruction: input.instructionHint,
+        target_artifact_types: input.targetArtifactTypes,
+        golden_path: `eval/goldens/${id}/`,
+        golden_pending: true,
+        ignore: ['**/@Id', '**/ModelSaveInfo'],
+        tags: input.tags ?? [],
+        split: 'holdout',
+    };
+}
+/**
+ * Write a mined case to eval/cases/<id>.json. Refuses to overwrite an existing
+ * case (mining should never silently clobber a hand-authored or already-mined
+ * case — pick a different idSlug, or edit the file directly).
+ */
+export function writeMinedCase(spec, casesDir) {
+    const outFile = path.join(casesDir, `${spec.id}.json`);
+    if (fs.existsSync(outFile)) {
+        throw new Error(`${outFile} already exists — mining refuses to overwrite. Pick a different idSlug.`);
+    }
+    fs.writeFileSync(outFile, JSON.stringify(spec, null, 2) + '\n');
+    return outFile;
+}
+//# sourceMappingURL=caseMining.js.map

package/dist/eval/improver/caseMining.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"caseMining.js","sourceRoot":"","sources":["../../../src/eval/improver/caseMining.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,KAAK,EAAE,MAAM,IAAI,CAAC;AACzB,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AA6B7B,2FAA2F;AAC3F,MAAM,UAAU,OAAO,CAAC,IAAY;IAClC,OAAO,IAAI;SACR,WAAW,EAAE;SACb,OAAO,CAAC,aAAa,EAAE,GAAG,CAAC;SAC3B,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC;SACvB,OAAO,CAAC,QAAQ,EAAE,GAAG,CAAC;SACtB,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AAClB,CAAC;AAED,qFAAqF;AACrF,MAAM,UAAU,mBAAmB,CAAC,KAAwB;IAC1D,MAAM,IAAI,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IACzE,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACpB,MAAM,IAAI,KAAK,CAAC,gDAAgD,KAAK,CAAC,KAAK,6BAA6B,CAAC,CAAC;IAC5G,CAAC;IACD,MAAM,EAAE,GAAG,IAAI,KAAK,CAAC,IAAI,IAAI,IAAI,EAAE,CAAC;IACpC,OAAO;QACL,EAAE;QACF,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,WAAW,EAAE,KAAK,CAAC,eAAe;QAClC,qBAAqB,EAAE,KAAK,CAAC,mBAAmB;QAChD,WAAW,EAAE,gBAAgB,EAAE,GAAG;QAClC,cAAc,EAAE,IAAI;QACpB,MAAM,EAAE,CAAC,QAAQ,EAAE,kBAAkB,CAAC;QACtC,IAAI,EAAE,KAAK,CAAC,IAAI,IAAI,EAAE;QACtB,KAAK,EAAE,SAAS;KACjB,CAAC;AACJ,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,cAAc,CAAC,IAAmB,EAAE,QAAgB;IAClE,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,EAAE,OAAO,CAAC,CAAC;IACvD,IAAI,EAAE,CAAC,UAAU,CAAC,OAAO,CAAC,EAAE,CAAC;QAC3B,MAAM,IAAI,KAAK,CAAC,GAAG,OAAO,yEAAyE,CAAC,CAAC;IACvG,CAAC;IACD,EAAE,CAAC,aAAa,CAAC,OAAO,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC;IAChE,OAAO,OAAO,CAAC;AACjB,CAAC"}

package/dist/eval/improver/caseMiningCli.d.ts

@@ -0,0 +1,10 @@
+/**
+ * Case mining CLI — turn a real failure description into a draft eval case
+ * (docs/AGENT_EVAL_LOOP.md §9). VM-free; writes eval/cases/<id>.json.
+ *
+ *   tsx src/eval/improver/caseMiningCli.ts \
+ *     --title "..." --tier 2 --instruction "..." --types AxClass,AxTable \
+ *     [--tags coc,extension] [--id custom-slug] [--dry-run]
+ */
+export {};
+//# sourceMappingURL=caseMiningCli.d.ts.map

package/dist/eval/improver/caseMiningCli.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"caseMiningCli.d.ts","sourceRoot":"","sources":["../../../src/eval/improver/caseMiningCli.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG"}

package/dist/eval/improver/caseMiningCli.js

@@ -0,0 +1,50 @@
+/**
+ * Case mining CLI — turn a real failure description into a draft eval case
+ * (docs/AGENT_EVAL_LOOP.md §9). VM-free; writes eval/cases/<id>.json.
+ *
+ *   tsx src/eval/improver/caseMiningCli.ts \
+ *     --title "..." --tier 2 --instruction "..." --types AxClass,AxTable \
+ *     [--tags coc,extension] [--id custom-slug] [--dry-run]
+ */
+import * as path from 'path';
+import { fileURLToPath } from 'url';
+import { mineCaseFromFailure, writeMinedCase } from './caseMining.js';
+const HERE = path.dirname(fileURLToPath(import.meta.url));
+const REPO_ROOT = path.resolve(HERE, '..', '..', '..');
+function arg(flag) {
+    const i = process.argv.indexOf(flag);
+    return i >= 0 ? process.argv[i + 1] : undefined;
+}
+function flagSet(flag) {
+    return process.argv.includes(flag);
+}
+function main() {
+    const title = arg('--title');
+    const tier = arg('--tier');
+    const instruction = arg('--instruction');
+    const types = arg('--types');
+    if (!title || !tier || !instruction || !types) {
+        console.error('usage: tsx src/eval/improver/caseMiningCli.ts --title "..." --tier N --instruction "..." ' +
+            '--types AxClass,AxTable [--tags a,b] [--id custom-slug] [--dry-run]');
+        process.exit(2);
+    }
+    const input = {
+        title,
+        tier: Number(tier),
+        instructionHint: instruction,
+        targetArtifactTypes: types.split(',').map(s => s.trim()).filter(Boolean),
+        tags: arg('--tags')?.split(',').map(s => s.trim()).filter(Boolean),
+        idSlug: arg('--id'),
+    };
+    const spec = mineCaseFromFailure(input);
+    if (flagSet('--dry-run')) {
+        console.log(JSON.stringify(spec, null, 2));
+        return;
+    }
+    const casesDir = path.join(REPO_ROOT, 'eval', 'cases');
+    const outFile = writeMinedCase(spec, casesDir);
+    console.log(`Wrote draft case: ${path.relative(REPO_ROOT, outFile)}`);
+    console.log('Next: refine `instruction` for precise reproducibility, then run it on the VM to capture the golden (§6.4) and drop golden_pending.');
+}
+main();
+//# sourceMappingURL=caseMiningCli.js.map

package/dist/eval/improver/caseMiningCli.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"caseMiningCli.js","sourceRoot":"","sources":["../../../src/eval/improver/caseMiningCli.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,EAAE,aAAa,EAAE,MAAM,KAAK,CAAC;AACpC,OAAO,EAAE,mBAAmB,EAAE,cAAc,EAA0B,MAAM,iBAAiB,CAAC;AAE9F,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAC1D,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;AAEvD,SAAS,GAAG,CAAC,IAAY;IACvB,MAAM,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IACrC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAClD,CAAC;AACD,SAAS,OAAO,CAAC,IAAY;IAC3B,OAAO,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;AACrC,CAAC;AAED,SAAS,IAAI;IACX,MAAM,KAAK,GAAG,GAAG,CAAC,SAAS,CAAC,CAAC;IAC7B,MAAM,IAAI,GAAG,GAAG,CAAC,QAAQ,CAAC,CAAC;IAC3B,MAAM,WAAW,GAAG,GAAG,CAAC,eAAe,CAAC,CAAC;IACzC,MAAM,KAAK,GAAG,GAAG,CAAC,SAAS,CAAC,CAAC;IAE7B,IAAI,CAAC,KAAK,IAAI,CAAC,IAAI,IAAI,CAAC,WAAW,IAAI,CAAC,KAAK,EAAE,CAAC;QAC9C,OAAO,CAAC,KAAK,CACX,2FAA2F;YAC3F,qEAAqE,CACtE,CAAC;QACF,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IAED,MAAM,KAAK,GAAsB;QAC/B,KAAK;QACL,IAAI,EAAE,MAAM,CAAC,IAAI,CAA8B;QAC/C,eAAe,EAAE,WAAW;QAC5B,mBAAmB,EAAE,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC;QACxE,IAAI,EAAE,GAAG,CAAC,QAAQ,CAAC,EAAE,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC;QAClE,MAAM,EAAE,GAAG,CAAC,MAAM,CAAC;KACpB,CAAC;IAEF,MAAM,IAAI,GAAG,mBAAmB,CAAC,KAAK,CAAC,CAAC;IAExC,IAAI,OAAO,CAAC,WAAW,CAAC,EAAE,CAAC;QACzB,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;QAC3C,OAAO;IACT,CAAC;IAED,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;IACvD,MAAM,OAAO,GAAG,cAAc,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;IAC/C,OAAO,CAAC,GAAG,CAAC,qBAAqB,IAAI,CAAC,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC,EAAE,CAAC,CAAC;IACtE,OAAO,CAAC,GAAG,CAAC,qIAAqI,CAAC,CAAC;AACrJ,CAAC;AAED,IAAI,EAAE,CAAC"}

package/dist/eval/improver/cli.d.ts

@@ -0,0 +1,11 @@
+/**
+ * Improver CLI — summarise the corpus into ranked failure clusters.
+ *
+ *   tsx src/eval/improver/cli.ts [--all] [--json]
+ *
+ * Reads every eval/corpus/runs/*.json and prints the actionable
+ * (TOOL_DEFECT/KNOWLEDGE_GAP/VALIDATOR_GAP) clusters ranked by frequency ×
+ * tier_weight — the input a human/agent uses to pick the next fix.
+ */
+export {};
+//# sourceMappingURL=cli.d.ts.map

package/dist/eval/improver/cli.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli.d.ts","sourceRoot":"","sources":["../../../src/eval/improver/cli.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG"}

package/dist/eval/improver/cli.js

@@ -0,0 +1,73 @@
+/**
+ * Improver CLI — summarise the corpus into ranked failure clusters.
+ *
+ *   tsx src/eval/improver/cli.ts [--all] [--json]
+ *
+ * Reads every eval/corpus/runs/*.json and prints the actionable
+ * (TOOL_DEFECT/KNOWLEDGE_GAP/VALIDATOR_GAP) clusters ranked by frequency ×
+ * tier_weight — the input a human/agent uses to pick the next fix.
+ */
+import * as fs from 'fs';
+import * as path from 'path';
+import { fileURLToPath } from 'url';
+import { clusterRuns, renderClusters } from './cluster.js';
+import { loadJsonRecords } from './corpusIO.js';
+import { aggregateBySplit, renderSplitReport, } from './heldout.js';
+const HERE = path.dirname(fileURLToPath(import.meta.url));
+const REPO_ROOT = path.resolve(HERE, '..', '..', '..');
+function loadRuns() {
+    const dir = path.join(REPO_ROOT, 'eval', 'corpus', 'runs');
+    return loadJsonRecords(dir, (r) => r != null && typeof r === 'object' && typeof r.classification === 'string');
+}
+/** Map case id → split, read from the case specs (default holdout). */
+function caseSplits() {
+    const dir = path.join(REPO_ROOT, 'eval', 'cases');
+    const m = new Map();
+    if (!fs.existsSync(dir))
+        return m;
+    for (const f of fs.readdirSync(dir)) {
+        if (!f.endsWith('.json'))
+            continue;
+        try {
+            const spec = JSON.parse(fs.readFileSync(path.join(dir, f), 'utf8'));
+            if (spec.id)
+                m.set(spec.id, spec.split === 'train' ? 'train' : 'holdout');
+        }
+        catch { /* skip */ }
+    }
+    return m;
+}
+/** Latest run per case → ScoredCase (split-aware) for the scoreboard. */
+function latestScoredCases(runs, splits) {
+    const latest = new Map();
+    for (const r of runs) {
+        const prev = latest.get(r.case_id);
+        if (!prev || r.run_id > prev.run_id)
+            latest.set(r.case_id, r);
+    }
+    return [...latest.values()].map(r => ({
+        caseId: r.case_id,
+        split: splits.get(r.case_id) ?? 'holdout',
+        score: {
+            build: r.score?.build ?? 0,
+            bp_clean: r.score?.bp_clean ?? 0,
+            golden_match: r.score?.golden_match ?? 0,
+        },
+    }));
+}
+const includeAll = process.argv.includes('--all');
+const asJson = process.argv.includes('--json');
+const runs = loadRuns();
+const clusters = clusterRuns(runs, includeAll);
+const scored = latestScoredCases(runs, caseSplits());
+const splitAgg = aggregateBySplit(scored);
+if (asJson) {
+    console.log(JSON.stringify({ clusters, splitScores: splitAgg }, null, 2));
+}
+else {
+    console.log(`Loaded ${runs.length} corpus run(s).\n`);
+    console.log(renderSplitReport(splitAgg));
+    console.log('');
+    console.log(renderClusters(clusters));
+}
+//# sourceMappingURL=cli.js.map

package/dist/eval/improver/cli.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli.js","sourceRoot":"","sources":["../../../src/eval/improver/cli.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,KAAK,EAAE,MAAM,IAAI,CAAC;AACzB,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,EAAE,aAAa,EAAE,MAAM,KAAK,CAAC;AACpC,OAAO,EAAE,WAAW,EAAE,cAAc,EAAkB,MAAM,cAAc,CAAC;AAC3E,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAChD,OAAO,EACL,gBAAgB,EAAE,iBAAiB,GACpC,MAAM,cAAc,CAAC;AAEtB,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAC1D,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;AAMvD,SAAS,QAAQ;IACf,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;IAC3D,OAAO,eAAe,CACpB,GAAG,EACH,CAAC,CAAC,EAAsB,EAAE,CAAC,CAAC,IAAI,IAAI,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,OAAQ,CAAmB,CAAC,cAAc,KAAK,QAAQ,CACzH,CAAC;AACJ,CAAC;AAED,uEAAuE;AACvE,SAAS,UAAU;IACjB,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;IAClD,MAAM,CAAC,GAAG,IAAI,GAAG,EAAiB,CAAC;IACnC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC;QAAE,OAAO,CAAC,CAAC;IAClC,KAAK,MAAM,CAAC,IAAI,EAAE,CAAC,WAAW,CAAC,GAAG,CAAC,EAAE,CAAC;QACpC,IAAI,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC;YAAE,SAAS;QACnC,IAAI,CAAC;YACH,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;YACpE,IAAI,IAAI,CAAC,EAAE;gBAAE,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,EAAE,IAAI,CAAC,KAAK,KAAK,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;QAC5E,CAAC;QAAC,MAAM,CAAC,CAAC,UAAU,CAAC,CAAC;IACxB,CAAC;IACD,OAAO,CAAC,CAAC;AACX,CAAC;AAED,yEAAyE;AACzE,SAAS,iBAAiB,CAAC,IAAqB,EAAE,MAA0B;IAC1E,MAAM,MAAM,GAAG,IAAI,GAAG,EAAyB,CAAC;IAChD,KAAK,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC;QACrB,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;QACnC,IAAI,CAAC,IAAI,IAAI,CAAC,CAAC,MAAM,GAAG,IAAI,CAAC,MAAM;YAAE,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QACpC,MAAM,EAAE,CAAC,CAAC,OAAO;QACjB,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,SAAS;QACzC,KAAK,EAAE;YACL,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,KAAK,IAAI,CAAC;YAC1B,QAAQ,EAAE,CAAC,CAAC,KAAK,EAAE,QAAQ,IAAI,CAAC;YAChC,YAAY,EAAE,CAAC,CAAC,KAAK,EAAE,YAAY,IAAI,CAAC;SACzC;KACF,CAAC,CAAC,CAAC;AACN,CAAC;AAED,MAAM,UAAU,GAAG,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;AAClD,MAAM,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;AAC/C,MAAM,IAAI,GAAG,QAAQ,EAAE,CAAC;AACxB,MAAM,QAAQ,GAAG,WAAW,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC;AAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,EAAE,UAAU,EAAE,CAAC,CAAC;AACrD,MAAM,QAAQ,GAAG,gBAAgB,CAAC,MAAM,CAAC,CAAC;AAE1C,IAAI,MAAM,EAAE,CAAC;IACX,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,QAAQ,EAAE,WAAW,EAAE,QAAQ,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;AAC5E,CAAC;KAAM,CAAC;IACN,OAAO,CAAC,GAAG,CAAC,UAAU,IAAI,CAAC,MAAM,mBAAmB,CAAC,CAAC;IACtD,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,QAAQ,CAAC,CAAC,CAAC;IACzC,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAChB,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,QAAQ,CAAC,CAAC,CAAC;AACxC,CAAC"}

package/dist/eval/improver/cluster.d.ts

@@ -0,0 +1,47 @@
+/**
+ * Corpus clustering for the improver (docs/AGENT_EVAL_LOOP.md §10.1–10.2).
+ *
+ * Groups non-PASS corpus runs by (classification, symptom) and ranks the clusters
+ * by `frequency × tier_weight` so the highest-leverage tool/knowledge/validator
+ * gap is picked first. Pure + VM-free: the CLI feeds it records read from disk.
+ */
+export interface CorpusRun {
+    run_id: string;
+    case_id: string;
+    tier: number;
+    classification: string;
+    root_cause_hypothesis?: string;
+    suggested_fix_area?: string;
+    score?: {
+        tier_weight?: number;
+        build?: number;
+        bp_clean?: number;
+        golden_match?: number;
+    };
+}
+export interface Cluster {
+    classification: string;
+    symptom: string;
+    /** Number of runs in the cluster. */
+    frequency: number;
+    /** Sum of tier weights (severity proxy). */
+    tierWeightSum: number;
+    /** Ranking score = frequency × max tier weight in the cluster. */
+    priority: number;
+    caseIds: string[];
+    runIds: string[];
+}
+/**
+ * Derive a short, stable symptom key for a run. Clusters group FAILURES, so the
+ * root-cause description (what went wrong) is a more stable key than the fix area
+ * (which may read "FIXED: …"). Fall back to fix area, then the classification.
+ */
+export declare function symptomOf(run: CorpusRun): string;
+/**
+ * Cluster the actionable (non-PASS) runs. `includeAll=true` keeps every class
+ * (incl. PASS / MODEL_ERROR / ENV_FLAKE) — useful for reporting, not prioritising.
+ */
+export declare function clusterRuns(runs: CorpusRun[], includeAll?: boolean): Cluster[];
+/** Render ranked clusters as a short report. */
+export declare function renderClusters(clusters: Cluster[]): string;
+//# sourceMappingURL=cluster.d.ts.map

package/dist/eval/improver/cluster.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cluster.d.ts","sourceRoot":"","sources":["../../../src/eval/improver/cluster.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,MAAM,WAAW,SAAS;IACxB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,cAAc,EAAE,MAAM,CAAC;IACvB,qBAAqB,CAAC,EAAE,MAAM,CAAC;IAC/B,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,KAAK,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,EAAE,MAAM,CAAC;QAAC,YAAY,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC;CAC5F;AAED,MAAM,WAAW,OAAO;IACtB,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,qCAAqC;IACrC,SAAS,EAAE,MAAM,CAAC;IAClB,4CAA4C;IAC5C,aAAa,EAAE,MAAM,CAAC;IACtB,kEAAkE;IAClE,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,MAAM,EAAE,MAAM,EAAE,CAAC;CAClB;AAcD;;;;GAIG;AACH,wBAAgB,SAAS,CAAC,GAAG,EAAE,SAAS,GAAG,MAAM,CAQhD;AAED;;;GAGG;AACH,wBAAgB,WAAW,CAAC,IAAI,EAAE,SAAS,EAAE,EAAE,UAAU,UAAQ,GAAG,OAAO,EAAE,CA0C5E;AAED,gDAAgD;AAChD,wBAAgB,cAAc,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG,MAAM,CAS1D"}

package/dist/eval/improver/cluster.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cluster.js","sourceRoot":"","sources":["../../../src/eval/improver/cluster.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAyBH,+EAA+E;AAC/E,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,CAAC,aAAa,EAAE,eAAe,EAAE,eAAe,CAAC,CAAC,CAAC;AAE9E,kFAAkF;AAClF,SAAS,WAAW,CAAC,IAAY;IAC/B,MAAM,OAAO,GAAG,IAAI;SACjB,OAAO,CAAC,gDAAgD,EAAE,EAAE,CAAC;SAC7D,OAAO,CAAC,yBAAyB,EAAE,EAAE,CAAC;SACtC,IAAI,EAAE,CAAC;IACV,OAAO,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AACtD,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,SAAS,CAAC,GAAc;IACtC,KAAK,MAAM,KAAK,IAAI,CAAC,GAAG,CAAC,qBAAqB,EAAE,GAAG,CAAC,kBAAkB,CAAC,EAAE,CAAC;QACxE,IAAI,KAAK,IAAI,KAAK,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YACjC,MAAM,MAAM,GAAG,WAAW,CAAC,KAAK,CAAC,CAAC;YAClC,IAAI,MAAM,CAAC,MAAM,IAAI,CAAC;gBAAE,OAAO,MAAM,CAAC;QACxC,CAAC;IACH,CAAC;IACD,OAAO,GAAG,CAAC,cAAc,CAAC;AAC5B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,WAAW,CAAC,IAAiB,EAAE,UAAU,GAAG,KAAK;IAC/D,MAAM,OAAO,GAAG,IAAI,GAAG,EAA+C,CAAC;IAEvE,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,IAAI,CAAC,UAAU,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,GAAG,CAAC,cAAc,CAAC;YAAE,SAAS;QACjE,MAAM,OAAO,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC;QAC/B,MAAM,GAAG,GAAG,GAAG,GAAG,CAAC,cAAc,IAAI,OAAO,EAAE,CAAC;QAC/C,MAAM,EAAE,GAAG,GAAG,CAAC,KAAK,EAAE,WAAW,IAAI,GAAG,CAAC,IAAI,IAAI,CAAC,CAAC;QACnD,IAAI,CAAC,GAAG,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACzB,IAAI,CAAC,CAAC,EAAE,CAAC;YACP,CAAC,GAAG;gBACF,cAAc,EAAE,GAAG,CAAC,cAAc;gBAClC,OAAO;gBACP,SAAS,EAAE,CAAC;gBACZ,aAAa,EAAE,CAAC;gBAChB,QAAQ,EAAE,CAAC;gBACX,OAAO,EAAE,EAAE;gBACX,MAAM,EAAE,EAAE;gBACV,WAAW,EAAE,EAAE;aAChB,CAAC;YACF,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;QACtB,CAAC;QACD,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC;QACjB,CAAC,CAAC,aAAa,IAAI,EAAE,CAAC;QACtB,CAAC,CAAC,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACvB,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC;YAAE,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAClE,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;IAC5B,CAAC;IAED,MAAM,QAAQ,GAAc,CAAC,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE;QACxD,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,WAAW,CAAC,CAAC;QAC5C,MAAM,EAAE,WAAW,EAAE,GAAG,IAAI,EAAE,GAAG,CAAC,CAAC;QACnC,OAAO,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,SAAS,GAAG,KAAK,EAAE,CAAC;IACpD,CAAC,CAAC,CAAC;IAEH,wEAAwE;IACxE,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CACrB,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,QAAQ;QACvB,CAAC,CAAC,SAAS,GAAG,CAAC,CAAC,SAAS;QACzB,CAAC,CAAC,cAAc,CAAC,aAAa,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC;IAEpD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,gDAAgD;AAChD,MAAM,UAAU,cAAc,CAAC,QAAmB;IAChD,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,sDAAsD,CAAC;IACzF,MAAM,KAAK,GAAa,CAAC,8BAA8B,QAAQ,CAAC,MAAM,yBAAyB,CAAC,CAAC;IACjG,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACxB,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,cAAc,KAAK,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;QAC3D,KAAK,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC,QAAQ,UAAU,CAAC,CAAC,SAAS,WAAW,CAAC,CAAC,aAAa,WAAW,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACtH,KAAK,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAChD,CAAC,CAAC,CAAC;IACH,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC"}

package/dist/eval/improver/corpusIO.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Shared corpus-record loading for the improver CLIs (docs/AGENT_EVAL_LOOP.md
+ * §10). VM-free — reads `eval/corpus/runs/*.json` from disk.
+ *
+ * The implementer (running on Windows/PowerShell) writes some corpus records
+ * with a UTF-8 BOM. `JSON.parse` throws on a leading BOM, and every loader in
+ * this package swallows parse errors in a bare `catch { return null }` so a
+ * malformed run is skipped rather than crashing the CLI — which silently
+ * dropped the majority of the corpus (52/60 files observed 2026-07-08) from
+ * every report/cluster/brief/flake/knowledge CLI without any error surfaced.
+ * Strip the BOM before parsing so a valid-JSON-except-for-BOM file loads like
+ * any other record.
+ */
+/** Strip a leading UTF-8 BOM (U+FEFF), if present, from file text. */
+export declare function stripBom(text: string): string;
+/** Parse a JSON file, tolerating a leading UTF-8 BOM. */
+export declare function readJsonLenient<T = unknown>(filePath: string): T;
+/**
+ * Read every `*.json` file directly under `dir`, parsing each with
+ * {@link readJsonLenient} and keeping only records that pass `isValid`.
+ * A file that fails to parse (bad JSON, not just BOM) or fails `isValid` is
+ * silently skipped — callers that need to know about skips should check the
+ * returned array length against the directory listing themselves.
+ */
+export declare function loadJsonRecords<T>(dir: string, isValid: (r: unknown) => r is T): T[];
+//# sourceMappingURL=corpusIO.d.ts.map

package/dist/eval/improver/corpusIO.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"corpusIO.d.ts","sourceRoot":"","sources":["../../../src/eval/improver/corpusIO.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAKH,sEAAsE;AACtE,wBAAgB,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAE7C;AAED,yDAAyD;AACzD,wBAAgB,eAAe,CAAC,CAAC,GAAG,OAAO,EAAE,QAAQ,EAAE,MAAM,GAAG,CAAC,CAEhE;AAED;;;;;;GAMG;AACH,wBAAgB,eAAe,CAAC,CAAC,EAAE,GAAG,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC,CAAC,EAAE,OAAO,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAcpF"}

package/dist/eval/improver/corpusIO.js

@@ -0,0 +1,50 @@
+/**
+ * Shared corpus-record loading for the improver CLIs (docs/AGENT_EVAL_LOOP.md
+ * §10). VM-free — reads `eval/corpus/runs/*.json` from disk.
+ *
+ * The implementer (running on Windows/PowerShell) writes some corpus records
+ * with a UTF-8 BOM. `JSON.parse` throws on a leading BOM, and every loader in
+ * this package swallows parse errors in a bare `catch { return null }` so a
+ * malformed run is skipped rather than crashing the CLI — which silently
+ * dropped the majority of the corpus (52/60 files observed 2026-07-08) from
+ * every report/cluster/brief/flake/knowledge CLI without any error surfaced.
+ * Strip the BOM before parsing so a valid-JSON-except-for-BOM file loads like
+ * any other record.
+ */
+import * as fs from 'fs';
+import * as path from 'path';
+/** Strip a leading UTF-8 BOM (U+FEFF), if present, from file text. */
+export function stripBom(text) {
+    return text.charCodeAt(0) === 0xfeff ? text.slice(1) : text;
+}
+/** Parse a JSON file, tolerating a leading UTF-8 BOM. */
+export function readJsonLenient(filePath) {
+    return JSON.parse(stripBom(fs.readFileSync(filePath, 'utf8')));
+}
+/**
+ * Read every `*.json` file directly under `dir`, parsing each with
+ * {@link readJsonLenient} and keeping only records that pass `isValid`.
+ * A file that fails to parse (bad JSON, not just BOM) or fails `isValid` is
+ * silently skipped — callers that need to know about skips should check the
+ * returned array length against the directory listing themselves.
+ */
+export function loadJsonRecords(dir, isValid) {
+    if (!fs.existsSync(dir))
+        return [];
+    const out = [];
+    for (const f of fs.readdirSync(dir)) {
+        if (!f.endsWith('.json'))
+            continue;
+        let parsed;
+        try {
+            parsed = readJsonLenient(path.join(dir, f));
+        }
+        catch {
+            continue;
+        }
+        if (isValid(parsed))
+            out.push(parsed);
+    }
+    return out;
+}
+//# sourceMappingURL=corpusIO.js.map

package/dist/eval/improver/corpusIO.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"corpusIO.js","sourceRoot":"","sources":["../../../src/eval/improver/corpusIO.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,KAAK,EAAE,MAAM,IAAI,CAAC;AACzB,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAE7B,sEAAsE;AACtE,MAAM,UAAU,QAAQ,CAAC,IAAY;IACnC,OAAO,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;AAC9D,CAAC;AAED,yDAAyD;AACzD,MAAM,UAAU,eAAe,CAAc,QAAgB;IAC3D,OAAO,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC,YAAY,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAM,CAAC;AACtE,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,eAAe,CAAI,GAAW,EAAE,OAA+B;IAC7E,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC;QAAE,OAAO,EAAE,CAAC;IACnC,MAAM,GAAG,GAAQ,EAAE,CAAC;IACpB,KAAK,MAAM,CAAC,IAAI,EAAE,CAAC,WAAW,CAAC,GAAG,CAAC,EAAE,CAAC;QACpC,IAAI,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC;YAAE,SAAS;QACnC,IAAI,MAAe,CAAC;QACpB,IAAI,CAAC;YACH,MAAM,GAAG,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC;QAC9C,CAAC;QAAC,MAAM,CAAC;YACP,SAAS;QACX,CAAC;QACD,IAAI,OAAO,CAAC,MAAM,CAAC;YAAE,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACxC,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC"}

package/dist/eval/improver/fixBrief.d.ts

@@ -0,0 +1,34 @@
+/**
+ * Fix brief generator (docs/AGENT_EVAL_LOOP.md §10, "Autonomous improver"). VM-free.
+ *
+ * Turns the TOP-PRIORITY actionable cluster (from cluster.ts) into a
+ * self-contained Markdown brief: symptom, root cause, evidence, and a concrete
+ * task list. This is the hand-off artifact between the two halves of the
+ * autonomous-improver idea:
+ *
+ *   VM half (this repo, interactive session)      VM-free half (CI runner)
+ *   ─────────────────────────────────────────     ──────────────────────────
+ *   run cases on the D365FO VM → corpus records    reproduce as a failing repo
+ *   → `npm run eval:brief` picks the top   ──brief──▶  test → fix → `npx vitest
+ *   cluster and writes this brief                  run` + held-out gate → PR
+ *
+ * The corpus (`eval/corpus/runs/`) is gitignored — it never reaches a CI
+ * runner's checkout. So a brief is the thing that DOES cross that boundary:
+ * generate it locally (where the corpus lives), then hand it to
+ * `.github/workflows/eval-improver.yml` (workflow_dispatch input) to run the
+ * reproduce→fix→test→PR loop, which needs only the repo + toolchain, not the
+ * VM. This module only builds the text; it never applies a fix itself.
+ */
+import { type CorpusRun, type Cluster } from './cluster.js';
+/** A corpus record with the evidence fields a brief needs (superset of CorpusRun). */
+export interface FixBriefRun extends CorpusRun {
+    timestamp?: string;
+    evidence_refs?: string[];
+}
+/** The single highest-priority actionable cluster, or null if the corpus is clean. */
+export declare function topPriorityCluster(runs: CorpusRun[]): Cluster | null;
+/** Render one cluster + its representative evidence as a self-contained Markdown brief. */
+export declare function renderFixBrief(cluster: Cluster, runs: FixBriefRun[]): string;
+/** Build the brief for the single top-priority cluster, or null if the corpus has nothing actionable. */
+export declare function buildTopFixBrief(runs: FixBriefRun[]): string | null;
+//# sourceMappingURL=fixBrief.d.ts.map

package/dist/eval/improver/fixBrief.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"fixBrief.d.ts","sourceRoot":"","sources":["../../../src/eval/improver/fixBrief.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,OAAO,EAAe,KAAK,SAAS,EAAE,KAAK,OAAO,EAAE,MAAM,cAAc,CAAC;AAEzE,sFAAsF;AACtF,MAAM,WAAW,WAAY,SAAQ,SAAS;IAC5C,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;CAC1B;AAaD,sFAAsF;AACtF,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,SAAS,EAAE,GAAG,OAAO,GAAG,IAAI,CAGpE;AAED,2FAA2F;AAC3F,wBAAgB,cAAc,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,GAAG,MAAM,CA4B5E;AAED,yGAAyG;AACzG,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,WAAW,EAAE,GAAG,MAAM,GAAG,IAAI,CAInE"}

package/dist/eval/improver/fixBrief.js

@@ -0,0 +1,69 @@
+/**
+ * Fix brief generator (docs/AGENT_EVAL_LOOP.md §10, "Autonomous improver"). VM-free.
+ *
+ * Turns the TOP-PRIORITY actionable cluster (from cluster.ts) into a
+ * self-contained Markdown brief: symptom, root cause, evidence, and a concrete
+ * task list. This is the hand-off artifact between the two halves of the
+ * autonomous-improver idea:
+ *
+ *   VM half (this repo, interactive session)      VM-free half (CI runner)
+ *   ─────────────────────────────────────────     ──────────────────────────
+ *   run cases on the D365FO VM → corpus records    reproduce as a failing repo
+ *   → `npm run eval:brief` picks the top   ──brief──▶  test → fix → `npx vitest
+ *   cluster and writes this brief                  run` + held-out gate → PR
+ *
+ * The corpus (`eval/corpus/runs/`) is gitignored — it never reaches a CI
+ * runner's checkout. So a brief is the thing that DOES cross that boundary:
+ * generate it locally (where the corpus lives), then hand it to
+ * `.github/workflows/eval-improver.yml` (workflow_dispatch input) to run the
+ * reproduce→fix→test→PR loop, which needs only the repo + toolchain, not the
+ * VM. This module only builds the text; it never applies a fix itself.
+ */
+import { clusterRuns } from './cluster.js';
+/** Pick the most informative run in a cluster: richest root_cause_hypothesis, most recent as tiebreak. */
+function representativeRun(cluster, runs) {
+    const inCluster = runs.filter(r => cluster.runIds.includes(r.run_id));
+    if (inCluster.length === 0)
+        return undefined;
+    return [...inCluster].sort((a, b) => {
+        const lenDiff = (b.root_cause_hypothesis?.length ?? 0) - (a.root_cause_hypothesis?.length ?? 0);
+        if (lenDiff !== 0)
+            return lenDiff;
+        return (b.timestamp ?? b.run_id).localeCompare(a.timestamp ?? a.run_id);
+    })[0];
+}
+/** The single highest-priority actionable cluster, or null if the corpus is clean. */
+export function topPriorityCluster(runs) {
+    const clusters = clusterRuns(runs, false);
+    return clusters[0] ?? null;
+}
+/** Render one cluster + its representative evidence as a self-contained Markdown brief. */
+export function renderFixBrief(cluster, runs) {
+    const rep = representativeRun(cluster, runs);
+    const lines = [
+        `# Fix brief: [${cluster.classification}] ${cluster.symptom}`,
+        '',
+        `Priority: ${cluster.priority}  (frequency=${cluster.frequency} × max tier_weight)`,
+        `Affected cases: ${cluster.caseIds.join(', ')}`,
+        `Corpus run(s): ${cluster.runIds.join(', ')}`,
+        '',
+        '## Root cause',
+        rep?.root_cause_hypothesis?.trim() || '(no root_cause_hypothesis recorded on the representative run)',
+        '',
+        '## Suggested fix area',
+        rep?.suggested_fix_area?.trim() || '(none recorded — investigate from the root cause above)',
+    ];
+    if (rep?.evidence_refs?.length) {
+        lines.push('', '## Evidence', ...rep.evidence_refs.map(e => `- ${e}`));
+    }
+    lines.push('', '## Task', '1. Reproduce this as a minimal FAILING test under `tests/` (mirroring the root cause above) — a real server-side gap must be demonstrable without the D365FO VM.', '2. Fix the root cause in `src/`.', '3. Run `npx vitest run` — the full suite must pass, including the new regression test.', '4. Run `npm run eval:clusters` against the corpus (if available) and confirm this symptom no longer ranks — or note that the corpus is VM-side evidence not present in this checkout, and the regression test is the durable proof instead.', '5. Open a PR citing this brief. Do NOT merge — a human reviews and merges.');
+    return lines.join('\n');
+}
+/** Build the brief for the single top-priority cluster, or null if the corpus has nothing actionable. */
+export function buildTopFixBrief(runs) {
+    const top = topPriorityCluster(runs);
+    if (!top)
+        return null;
+    return renderFixBrief(top, runs);
+}
+//# sourceMappingURL=fixBrief.js.map

package/dist/eval/improver/fixBrief.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"fixBrief.js","sourceRoot":"","sources":["../../../src/eval/improver/fixBrief.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,OAAO,EAAE,WAAW,EAAgC,MAAM,cAAc,CAAC;AAQzE,0GAA0G;AAC1G,SAAS,iBAAiB,CAAC,OAAgB,EAAE,IAAmB;IAC9D,MAAM,SAAS,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC;IACtE,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC7C,OAAO,CAAC,GAAG,SAAS,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QAClC,MAAM,OAAO,GAAG,CAAC,CAAC,CAAC,qBAAqB,EAAE,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,qBAAqB,EAAE,MAAM,IAAI,CAAC,CAAC,CAAC;QAChG,IAAI,OAAO,KAAK,CAAC;YAAE,OAAO,OAAO,CAAC;QAClC,OAAO,CAAC,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC;IAC1E,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACR,CAAC;AAED,sFAAsF;AACtF,MAAM,UAAU,kBAAkB,CAAC,IAAiB;IAClD,MAAM,QAAQ,GAAG,WAAW,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IAC1C,OAAO,QAAQ,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC;AAC7B,CAAC;AAED,2FAA2F;AAC3F,MAAM,UAAU,cAAc,CAAC,OAAgB,EAAE,IAAmB;IAClE,MAAM,GAAG,GAAG,iBAAiB,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC;IAC7C,MAAM,KAAK,GAAa;QACtB,iBAAiB,OAAO,CAAC,cAAc,KAAK,OAAO,CAAC,OAAO,EAAE;QAC7D,EAAE;QACF,aAAa,OAAO,CAAC,QAAQ,gBAAgB,OAAO,CAAC,SAAS,qBAAqB;QACnF,mBAAmB,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;QAC/C,kBAAkB,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;QAC7C,EAAE;QACF,eAAe;QACf,GAAG,EAAE,qBAAqB,EAAE,IAAI,EAAE,IAAI,+DAA+D;QACrG,EAAE;QACF,uBAAuB;QACvB,GAAG,EAAE,kBAAkB,EAAE,IAAI,EAAE,IAAI,yDAAyD;KAC7F,CAAC;IACF,IAAI,GAAG,EAAE,aAAa,EAAE,MAAM,EAAE,CAAC;QAC/B,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,aAAa,EAAE,GAAG,GAAG,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC;IACzE,CAAC;IACD,KAAK,CAAC,IAAI,CACR,EAAE,EACF,SAAS,EACT,kKAAkK,EAClK,kCAAkC,EAClC,wFAAwF,EACxF,6OAA6O,EAC7O,4EAA4E,CAC7E,CAAC;IACF,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,yGAAyG;AACzG,MAAM,UAAU,gBAAgB,CAAC,IAAmB;IAClD,MAAM,GAAG,GAAG,kBAAkB,CAAC,IAAI,CAAC,CAAC;IACrC,IAAI,CAAC,GAAG;QAAE,OAAO,IAAI,CAAC;IACtB,OAAO,cAAc,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;AACnC,CAAC"}

package/dist/eval/improver/fixBriefCli.d.ts

@@ -0,0 +1,11 @@
+/**
+ * Fix brief CLI — print (or write) the top-priority actionable cluster's fix
+ * brief, the hand-off artifact for the eval-improver GitHub Actions workflow
+ * (docs/AGENT_EVAL_LOOP.md §10). VM-free.
+ *
+ *   tsx src/eval/improver/fixBriefCli.ts [--out file.md] [--all]
+ *
+ * --all prints a brief for every actionable cluster, not just the top one.
+ */
+export {};
+//# sourceMappingURL=fixBriefCli.d.ts.map

package/dist/eval/improver/fixBriefCli.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"fixBriefCli.d.ts","sourceRoot":"","sources":["../../../src/eval/improver/fixBriefCli.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG"}

package/dist/eval/improver/fixBriefCli.js

@@ -0,0 +1,46 @@
+/**
+ * Fix brief CLI — print (or write) the top-priority actionable cluster's fix
+ * brief, the hand-off artifact for the eval-improver GitHub Actions workflow
+ * (docs/AGENT_EVAL_LOOP.md §10). VM-free.
+ *
+ *   tsx src/eval/improver/fixBriefCli.ts [--out file.md] [--all]
+ *
+ * --all prints a brief for every actionable cluster, not just the top one.
+ */
+import * as fs from 'fs';
+import * as path from 'path';
+import { fileURLToPath } from 'url';
+import { clusterRuns } from './cluster.js';
+import { renderFixBrief, buildTopFixBrief } from './fixBrief.js';
+import { loadJsonRecords } from './corpusIO.js';
+const HERE = path.dirname(fileURLToPath(import.meta.url));
+const REPO_ROOT = path.resolve(HERE, '..', '..', '..');
+function loadRuns() {
+    const dir = path.join(REPO_ROOT, 'eval', 'corpus', 'runs');
+    return loadJsonRecords(dir, (r) => r != null && typeof r === 'object' && typeof r.classification === 'string');
+}
+function arg(flag) {
+    const i = process.argv.indexOf(flag);
+    return i >= 0 ? process.argv[i + 1] : undefined;
+}
+const runs = loadRuns();
+const all = process.argv.includes('--all');
+let output;
+if (all) {
+    const clusters = clusterRuns(runs, false);
+    output = clusters.length === 0
+        ? 'No actionable failure clusters — corpus is clean. 🎉'
+        : clusters.map(c => renderFixBrief(c, runs)).join('\n\n---\n\n');
+}
+else {
+    output = buildTopFixBrief(runs) ?? 'No actionable failure clusters — corpus is clean. 🎉';
+}
+const outFile = arg('--out');
+if (outFile) {
+    fs.writeFileSync(outFile, output + '\n');
+    console.error(`Wrote brief to ${outFile}`);
+}
+else {
+    console.log(output);
+}
+//# sourceMappingURL=fixBriefCli.js.map

package/dist/eval/improver/fixBriefCli.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"fixBriefCli.js","sourceRoot":"","sources":["../../../src/eval/improver/fixBriefCli.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,KAAK,EAAE,MAAM,IAAI,CAAC;AACzB,OAAO,KAAK,IAAI,MAAM,MAAM,CAAC;AAC7B,OAAO,EAAE,aAAa,EAAE,MAAM,KAAK,CAAC;AACpC,OAAO,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAC3C,OAAO,EAAE,cAAc,EAAE,gBAAgB,EAAoB,MAAM,eAAe,CAAC;AACnF,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAEhD,MAAM,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAC1D,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;AAEvD,SAAS,QAAQ;IACf,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;IAC3D,OAAO,eAAe,CACpB,GAAG,EACH,CAAC,CAAC,EAAoB,EAAE,CAAC,CAAC,IAAI,IAAI,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,OAAQ,CAAiB,CAAC,cAAc,KAAK,QAAQ,CACrH,CAAC;AACJ,CAAC;AAED,SAAS,GAAG,CAAC,IAAY;IACvB,MAAM,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IACrC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAClD,CAAC;AAED,MAAM,IAAI,GAAG,QAAQ,EAAE,CAAC;AACxB,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;AAE3C,IAAI,MAAc,CAAC;AACnB,IAAI,GAAG,EAAE,CAAC;IACR,MAAM,QAAQ,GAAG,WAAW,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IAC1C,MAAM,GAAG,QAAQ,CAAC,MAAM,KAAK,CAAC;QAC5B,CAAC,CAAC,sDAAsD;QACxD,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,cAAc,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;AACrE,CAAC;KAAM,CAAC;IACN,MAAM,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,sDAAsD,CAAC;AAC5F,CAAC;AAED,MAAM,OAAO,GAAG,GAAG,CAAC,OAAO,CAAC,CAAC;AAC7B,IAAI,OAAO,EAAE,CAAC;IACZ,EAAE,CAAC,aAAa,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC,CAAC;IACzC,OAAO,CAAC,KAAK,CAAC,kBAAkB,OAAO,EAAE,CAAC,CAAC;AAC7C,CAAC;KAAM,CAAC;IACN,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;AACtB,CAAC"}

package/dist/eval/improver/flakeDetection.d.ts

@@ -0,0 +1,50 @@
+/**
+ * Flake detection (docs/AGENT_EVAL_LOOP.md §9, "Self-improving upgrades"). VM-free.
+ *
+ * A failure is only worth turning into a fix if it's REPRODUCIBLE (§9 rubric:
+ * "TOOL_DEFECT/VALIDATOR_GAP must be reproducible deterministically"). If two
+ * runs of the SAME case against the SAME server_git_sha (so no code changed
+ * between them — any difference can't be a real fix or regression) score
+ * differently, that disagreement is non-determinism: infra flake (build
+ * server hiccup, locking, timeout), not a real defect.
+ *
+ * This module only DETECTS and REPORTS candidates — it never rewrites a
+ * corpus record's classification. Corpus runs are immutable evidence keyed to
+ * server_git_sha (see eval/ROADMAP.md); a human/improver-agent reviews a flake
+ * candidate and, if confirmed, the NEXT run for that case gets classified
+ * ENV_FLAKE explicitly via the oracle CLI's --classification flag.
+ */
+/** Score fields compared for disagreement between same-sha runs. */
+declare const SCORE_FIELDS: readonly ['build', 'bp_clean', 'golden_match', 'systest'];
+type ScoreField = (typeof SCORE_FIELDS)[number];
+/**
+ * Minimal shape this module needs from a corpus record. Deliberately not
+ * `cluster.ts`'s `CorpusRun` — that type's `score` lacks `systest` and has no
+ * `server_git_sha`, neither of which fit a structural extension cleanly.
+ */
+export interface FlakeCorpusRun {
+    run_id: string;
+    case_id: string;
+    server_git_sha?: string;
+    score?: Partial<Record<ScoreField, number | null>>;
+}
+export interface ScoreDisagreement {
+    field: ScoreField;
+    /** run_id -> value, only for runs where the value differs from the group's first value. */
+    values: Record<string, number | null | undefined>;
+}
+export interface FlakeCandidate {
+    case_id: string;
+    server_git_sha: string;
+    runIds: string[];
+    disagreements: ScoreDisagreement[];
+}
+/**
+ * Group runs by (case_id, server_git_sha) and flag any group where the same
+ * case scored differently against the exact same code under test.
+ */
+export declare function detectFlakeCandidates(runs: FlakeCorpusRun[]): FlakeCandidate[];
+/** Render flake candidates as a short human-readable report. */
+export declare function renderFlakeCandidates(candidates: FlakeCandidate[]): string;
+export {};
+//# sourceMappingURL=flakeDetection.d.ts.map

package/dist/eval/improver/flakeDetection.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"flakeDetection.d.ts","sourceRoot":"","sources":["../../../src/eval/improver/flakeDetection.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,oEAAoE;AACpE,QAAA,MAAM,YAAY,YAAI,OAAO,EAAE,UAAU,EAAE,cAAc,EAAE,SAAS,CAAU,CAAC;AAC/E,KAAK,UAAU,GAAG,CAAC,OAAO,YAAY,CAAC,CAAC,MAAM,CAAC,CAAC;AAEhD;;;;GAIG;AACH,MAAM,WAAW,cAAc;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;IAChB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,KAAK,CAAC,EAAE,OAAO,CAAC,MAAM,CAAC,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC;CACpD;AAED,MAAM,WAAW,iBAAiB;IAChC,KAAK,EAAE,UAAU,CAAC;IAClB,2FAA2F;IAC3F,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC,CAAC;CACnD;AAED,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,cAAc,EAAE,MAAM,CAAC;IACvB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,aAAa,EAAE,iBAAiB,EAAE,CAAC;CACpC;AAED;;;GAGG;AACH,wBAAgB,qBAAqB,CAAC,IAAI,EAAE,cAAc,EAAE,GAAG,cAAc,EAAE,CAgC9E;AAED,gEAAgE;AAChE,wBAAgB,qBAAqB,CAAC,UAAU,EAAE,cAAc,EAAE,GAAG,MAAM,CAW1E"}