@prowi/deskcheck 0.1.0 → 0.3.0

package/build/services/testing/TestScorerService.js

@@ -0,0 +1,35 @@
+/**
+ * Calculate test scores from judge verdicts.
+ *
+ * Pure math — no LLM, no I/O. Computes recall, precision, and scope
+ * compliance from the judge's finding and expectation reviews.
+ */
+export function calculateScores(judgeResult) {
+    const { findings_review, expectations_review } = judgeResult;
+    // Finding counts by verdict
+    const total_findings = findings_review.length;
+    const correct = findings_review.filter((f) => f.verdict === "correct").length;
+    const out_of_scope = findings_review.filter((f) => f.verdict === "out_of_scope").length;
+    const incorrect_severity = findings_review.filter((f) => f.verdict === "incorrect_severity").length;
+    // Expectation counts by verdict
+    const total_expectations = expectations_review.length;
+    const expectations_found = expectations_review.filter((e) => e.verdict === "found").length;
+    const expectations_missed = expectations_review.filter((e) => e.verdict === "missed").length;
+    // Derived ratios (default to 1.0 when denominator is zero)
+    const recall = total_expectations > 0 ? expectations_found / total_expectations : 1.0;
+    const precision = total_findings > 0 ? correct / total_findings : 1.0;
+    const scope_compliance = total_findings > 0 ? (total_findings - out_of_scope) / total_findings : 1.0;
+    return {
+        total_findings,
+        correct,
+        out_of_scope,
+        incorrect_severity,
+        total_expectations,
+        expectations_found,
+        expectations_missed,
+        recall,
+        precision,
+        scope_compliance,
+    };
+}
+//# sourceMappingURL=TestScorerService.js.map

package/build/services/testing/TestScorerService.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"TestScorerService.js","sourceRoot":"","sources":["../../../src/services/testing/TestScorerService.ts"],"names":[],"mappings":"AAEA;;;;;GAKG;AACH,MAAM,UAAU,eAAe,CAAC,WAAwB;IACtD,MAAM,EAAE,eAAe,EAAE,mBAAmB,EAAE,GAAG,WAAW,CAAC;IAE7D,4BAA4B;IAC5B,MAAM,cAAc,GAAG,eAAe,CAAC,MAAM,CAAC;IAC9C,MAAM,OAAO,GAAG,eAAe,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,MAAM,CAAC;IAC9E,MAAM,YAAY,GAAG,eAAe,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,cAAc,CAAC,CAAC,MAAM,CAAC;IACxF,MAAM,kBAAkB,GAAG,eAAe,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,oBAAoB,CAAC,CAAC,MAAM,CAAC;IAEpG,gCAAgC;IAChC,MAAM,kBAAkB,GAAG,mBAAmB,CAAC,MAAM,CAAC;IACtD,MAAM,kBAAkB,GAAG,mBAAmB,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,OAAO,CAAC,CAAC,MAAM,CAAC;IAC3F,MAAM,mBAAmB,GAAG,mBAAmB,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,MAAM,CAAC;IAE7F,2DAA2D;IAC3D,MAAM,MAAM,GAAG,kBAAkB,GAAG,CAAC,CAAC,CAAC,CAAC,kBAAkB,GAAG,kBAAkB,CAAC,CAAC,CAAC,GAAG,CAAC;IACtF,MAAM,SAAS,GAAG,cAAc,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,GAAG,cAAc,CAAC,CAAC,CAAC,GAAG,CAAC;IACtE,MAAM,gBAAgB,GAAG,cAAc,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,cAAc,GAAG,YAAY,CAAC,GAAG,cAAc,CAAC,CAAC,CAAC,GAAG,CAAC;IAErG,OAAO;QACL,cAAc;QACd,OAAO;QACP,YAAY;QACZ,kBAAkB;QAClB,kBAAkB;QAClB,kBAAkB;QAClB,mBAAmB;QACnB,MAAM;QACN,SAAS;QACT,gBAAgB;KACjB,CAAC;AACJ,CAAC"}

package/build/services/testing/TestStorageService.d.ts

@@ -0,0 +1,39 @@
+import type { TestCase, TestCaseResult, TestRun, TestSuiteStatus } from "../../types/testing.js";
+/**
+ * Persists test runs to .deskcheck/test-runs/ using atomic writes.
+ *
+ * Each test run lives in a timestamped directory containing a single
+ * results.json file. Uses write-to-tmp + rename for atomic updates.
+ * No file locking needed since test runs execute sequentially.
+ */
+export declare class TestStorageService {
+    private readonly storageDir;
+    constructor(storageDir: string);
+    /**
+     * Create a new test run with all test cases initialized as "pending".
+     *
+     * Creates a timestamped directory and writes an initial results.json.
+     */
+    createRun(testCases: TestCase[]): TestRun;
+    /** Read the results.json for a given run ID. */
+    getRun(runId: string): TestRun;
+    /**
+     * Return the most recent run directory name, or null if no runs exist.
+     */
+    getLatestRunId(): string | null;
+    /**
+     * Update a specific test case in results.json.
+     *
+     * Reads the current run, applies the partial update to the specified
+     * test case, and writes back atomically.
+     */
+    updateTestCase(runId: string, criterionId: string, testName: string, update: Partial<TestCaseResult>): void;
+    /** Update a suite's status. */
+    updateSuiteStatus(runId: string, criterionId: string, status: TestSuiteStatus): void;
+    /** Mark the run as complete with a completion timestamp. */
+    completeRun(runId: string): void;
+    private resultsPath;
+    /** Atomic write: write to .tmp then rename. */
+    private writeRun;
+}
+//# sourceMappingURL=TestStorageService.d.ts.map

package/build/services/testing/TestStorageService.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"TestStorageService.d.ts","sourceRoot":"","sources":["../../../src/services/testing/TestStorageService.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EACV,QAAQ,EACR,cAAc,EACd,OAAO,EACP,eAAe,EAChB,MAAM,wBAAwB,CAAC;AAmBhC;;;;;;GAMG;AACH,qBAAa,kBAAkB;IAC7B,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAS;gBAExB,UAAU,EAAE,MAAM;IAI9B;;;;OAIG;IACH,SAAS,CAAC,SAAS,EAAE,QAAQ,EAAE,GAAG,OAAO;IA8CzC,gDAAgD;IAChD,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO;IAM9B;;OAEG;IACH,cAAc,IAAI,MAAM,GAAG,IAAI;IAsB/B;;;;;OAKG;IACH,cAAc,CACZ,KAAK,EAAE,MAAM,EACb,WAAW,EAAE,MAAM,EACnB,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,OAAO,CAAC,cAAc,CAAC,GAC9B,IAAI;IAiBP,+BAA+B;IAC/B,iBAAiB,CAAC,KAAK,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,MAAM,EAAE,eAAe,GAAG,IAAI;IAYpF,4DAA4D;IAC5D,WAAW,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI;IAWhC,OAAO,CAAC,WAAW;IAInB,+CAA+C;IAC/C,OAAO,CAAC,QAAQ;CAMjB"}

package/build/services/testing/TestStorageService.js

@@ -0,0 +1,144 @@
+import fs from "node:fs";
+import path from "node:path";
+// =============================================================================
+// Helpers
+// =============================================================================
+/** Format a Date as YYYY-MM-DD_HHmmss for use as a run ID / directory name. */
+function formatTimestamp(date) {
+    const pad2 = (n) => String(n).padStart(2, "0");
+    return (`${date.getFullYear()}-${pad2(date.getMonth() + 1)}-${pad2(date.getDate())}` +
+        `_${pad2(date.getHours())}${pad2(date.getMinutes())}${pad2(date.getSeconds())}`);
+}
+// =============================================================================
+// TestStorageService
+// =============================================================================
+/**
+ * Persists test runs to .deskcheck/test-runs/ using atomic writes.
+ *
+ * Each test run lives in a timestamped directory containing a single
+ * results.json file. Uses write-to-tmp + rename for atomic updates.
+ * No file locking needed since test runs execute sequentially.
+ */
+export class TestStorageService {
+    storageDir;
+    constructor(storageDir) {
+        this.storageDir = storageDir;
+    }
+    /**
+     * Create a new test run with all test cases initialized as "pending".
+     *
+     * Creates a timestamped directory and writes an initial results.json.
+     */
+    createRun(testCases) {
+        const now = new Date();
+        const runId = formatTimestamp(now);
+        const runDir = path.join(this.storageDir, runId);
+        fs.mkdirSync(runDir, { recursive: true });
+        // Group test cases by criterionId to build suites
+        const suites = {};
+        for (const testCase of testCases) {
+            if (!suites[testCase.criterionId]) {
+                suites[testCase.criterionId] = {
+                    status: "pending",
+                    criterion_file: testCase.criterionFile,
+                    tests: {},
+                };
+            }
+            const initialResult = {
+                status: "pending",
+                fixture_file: testCase.fixtureFile,
+                expected_file: testCase.expectedFile,
+                findings: null,
+                judge: null,
+                scores: null,
+                error: null,
+                executor_usage: null,
+                judge_usage: null,
+            };
+            suites[testCase.criterionId].tests[testCase.name] = initialResult;
+        }
+        const run = {
+            run_id: runId,
+            status: "running",
+            started_at: now.toISOString(),
+            completed_at: null,
+            suites,
+        };
+        this.writeRun(runId, run);
+        return run;
+    }
+    /** Read the results.json for a given run ID. */
+    getRun(runId) {
+        const resultsPath = this.resultsPath(runId);
+        const raw = fs.readFileSync(resultsPath, "utf-8");
+        return JSON.parse(raw);
+    }
+    /**
+     * Return the most recent run directory name, or null if no runs exist.
+     */
+    getLatestRunId() {
+        if (!fs.existsSync(this.storageDir)) {
+            return null;
+        }
+        const entries = fs.readdirSync(this.storageDir, { withFileTypes: true });
+        const runDirs = entries
+            .filter((entry) => entry.isDirectory() &&
+            fs.existsSync(path.join(this.storageDir, entry.name, "results.json")))
+            .map((entry) => entry.name)
+            .sort();
+        if (runDirs.length === 0) {
+            return null;
+        }
+        return runDirs[runDirs.length - 1];
+    }
+    /**
+     * Update a specific test case in results.json.
+     *
+     * Reads the current run, applies the partial update to the specified
+     * test case, and writes back atomically.
+     */
+    updateTestCase(runId, criterionId, testName, update) {
+        const run = this.getRun(runId);
+        const suite = run.suites[criterionId];
+        if (!suite) {
+            throw new Error(`Suite "${criterionId}" not found in run "${runId}"`);
+        }
+        const testCase = suite.tests[testName];
+        if (!testCase) {
+            throw new Error(`Test "${testName}" not found in suite "${criterionId}" of run "${runId}"`);
+        }
+        Object.assign(testCase, update);
+        this.writeRun(runId, run);
+    }
+    /** Update a suite's status. */
+    updateSuiteStatus(runId, criterionId, status) {
+        const run = this.getRun(runId);
+        const suite = run.suites[criterionId];
+        if (!suite) {
+            throw new Error(`Suite "${criterionId}" not found in run "${runId}"`);
+        }
+        suite.status = status;
+        this.writeRun(runId, run);
+    }
+    /** Mark the run as complete with a completion timestamp. */
+    completeRun(runId) {
+        const run = this.getRun(runId);
+        run.status = "complete";
+        run.completed_at = new Date().toISOString();
+        this.writeRun(runId, run);
+    }
+    // ---------------------------------------------------------------------------
+    // Private Helpers
+    // ---------------------------------------------------------------------------
+    resultsPath(runId) {
+        return path.join(this.storageDir, runId, "results.json");
+    }
+    /** Atomic write: write to .tmp then rename. */
+    writeRun(runId, run) {
+        const target = this.resultsPath(runId);
+        const tmp = target + ".tmp";
+        fs.writeFileSync(tmp, JSON.stringify(run, null, 2) + "\n");
+        fs.renameSync(tmp, target);
+    }
+}
+//# sourceMappingURL=TestStorageService.js.map

package/build/services/testing/TestStorageService.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"TestStorageService.js","sourceRoot":"","sources":["../../../src/services/testing/TestStorageService.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,SAAS,CAAC;AACzB,OAAO,IAAI,MAAM,WAAW,CAAC;AAQ7B,gFAAgF;AAChF,UAAU;AACV,gFAAgF;AAEhF,+EAA+E;AAC/E,SAAS,eAAe,CAAC,IAAU;IACjC,MAAM,IAAI,GAAG,CAAC,CAAS,EAAU,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IAC/D,OAAO,CACL,GAAG,IAAI,CAAC,WAAW,EAAE,IAAI,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC,EAAE;QAC5E,IAAI,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,CAAC,EAAE,CAChF,CAAC;AACJ,CAAC;AAED,gFAAgF;AAChF,qBAAqB;AACrB,gFAAgF;AAEhF;;;;;;GAMG;AACH,MAAM,OAAO,kBAAkB;IACZ,UAAU,CAAS;IAEpC,YAAY,UAAkB;QAC5B,IAAI,CAAC,UAAU,GAAG,UAAU,CAAC;IAC/B,CAAC;IAED;;;;OAIG;IACH,SAAS,CAAC,SAAqB;QAC7B,MAAM,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC;QACvB,MAAM,KAAK,GAAG,eAAe,CAAC,GAAG,CAAC,CAAC;QACnC,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC;QAEjD,EAAE,CAAC,SAAS,CAAC,MAAM,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAE1C,kDAAkD;QAClD,MAAM,MAAM,GAAsB,EAAE,CAAC;QAErC,KAAK,MAAM,QAAQ,IAAI,SAAS,EAAE,CAAC;YACjC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC;gBAClC,MAAM,CAAC,QAAQ,CAAC,WAAW,CAAC,GAAG;oBAC7B,MAAM,EAAE,SAAS;oBACjB,cAAc,EAAE,QAAQ,CAAC,aAAa;oBACtC,KAAK,EAAE,EAAE;iBACV,CAAC;YACJ,CAAC;YAED,MAAM,aAAa,GAAmB;gBACpC,MAAM,EAAE,SAAS;gBACjB,YAAY,EAAE,QAAQ,CAAC,WAAW;gBAClC,aAAa,EAAE,QAAQ,CAAC,YAAY;gBACpC,QAAQ,EAAE,IAAI;gBACd,KAAK,EAAE,IAAI;gBACX,MAAM,EAAE,IAAI;gBACZ,KAAK,EAAE,IAAI;gBACX,cAAc,EAAE,IAAI;gBACpB,WAAW,EAAE,IAAI;aAClB,CAAC;YAEF,MAAM,CAAC,QAAQ,CAAC,WAAW,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,aAAa,CAAC;QACpE,CAAC;QAED,MAAM,GAAG,GAAY;YACnB,MAAM,EAAE,KAAK;YACb,MAAM,EAAE,SAAS;YACjB,UAAU,EAAE,GAAG,CAAC,WAAW,EAAE;YAC7B,YAAY,EAAE,IAAI;YAClB,MAAM;SACP,CAAC;QAEF,IAAI,CAAC,QAAQ,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;QAC1B,OAAO,GAAG,CAAC;IACb,CAAC;IAED,gDAAgD;IAChD,MAAM,CAAC,KAAa;QAClB,MAAM,WAAW,GAAG,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QAC5C,MAAM,GAAG,GAAG,EAAE,CAAC,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;QAClD,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAY,CAAC;IACpC,CAAC;IAED;;OAEG;IACH,cAAc;QACZ,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,UAAU,CAAC,EAAE,CAAC;YACpC,OAAO,IAAI,CAAC;QACd,CAAC;QAED,MAAM,OAAO,GAAG,EAAE,CAAC,WAAW,CAAC,IAAI,CAAC,UAAU,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;QACzE,MAAM,OAAO,GAAG,OAAO;aACpB,MAAM,CACL,CAAC,KAAK,EAAE,EAAE,CACR,KAAK,CAAC,WAAW,EAAE;YACnB,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,KAAK,CAAC,IAAI,EAAE,cAAc,CAAC,CAAC,CACxE;aACA,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC;aAC1B,IAAI,EAAE,CAAC;QAEV,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACzB,OAAO,IAAI,CAAC;QACd,CAAC;QAED,OAAO,OAAO,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACrC,CAAC;IAED;;;;;OAKG;IACH,cAAc,CACZ,KAAa,EACb,WAAmB,EACnB,QAAgB,EAChB,MAA+B;QAE/B,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAC/B,MAAM,KAAK,GAAG,GAAG,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;QAEtC,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,MAAM,IAAI,KAAK,CAAC,UAAU,WAAW,uBAAuB,KAAK,GAAG,CAAC,CAAC;QACxE,CAAC;QAED,MAAM,QAAQ,GAAG,KAAK,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;QACvC,IAAI,CAAC,QAAQ,EAAE,CAAC;YACd,MAAM,IAAI,KAAK,CAAC,SAAS,QAAQ,yBAAyB,WAAW,aAAa,KAAK,GAAG,CAAC,CAAC;QAC9F,CAAC;QAED,MAAM,CAAC,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;QAChC,IAAI,CAAC,QAAQ,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IAC5B,CAAC;IAED,+BAA+B;IAC/B,iBAAiB,CAAC,KAAa,EAAE,WAAmB,EAAE,MAAuB;QAC3E,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAC/B,MAAM,KAAK,GAAG,GAAG,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;QAEtC,IAAI,CAAC,KAAK,EAAE,CAAC;YACX,MAAM,IAAI,KAAK,CAAC,UAAU,WAAW,uBAAuB,KAAK,GAAG,CAAC,CAAC;QACxE,CAAC;QAED,KAAK,CAAC,MAAM,GAAG,MAAM,CAAC;QACtB,IAAI,CAAC,QAAQ,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IAC5B,CAAC;IAED,4DAA4D;IAC5D,WAAW,CAAC,KAAa;QACvB,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;QAC/B,GAAG,CAAC,MAAM,GAAG,UAAU,CAAC;QACxB,GAAG,CAAC,YAAY,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QAC5C,IAAI,CAAC,QAAQ,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IAC5B,CAAC;IAED,8EAA8E;IAC9E,kBAAkB;IAClB,8EAA8E;IAEtE,WAAW,CAAC,KAAa;QAC/B,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,KAAK,EAAE,cAAc,CAAC,CAAC;IAC3D,CAAC;IAED,+CAA+C;IACvC,QAAQ,CAAC,KAAa,EAAE,GAAY;QAC1C,MAAM,MAAM,GAAG,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QACvC,MAAM,GAAG,GAAG,MAAM,GAAG,MAAM,CAAC;QAC5B,EAAE,CAAC,aAAa,CAAC,GAAG,EAAE,IAAI,CAAC,SAAS,CAAC,GAAG,EAAE,IAAI,EAAE,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC;QAC3D,EAAE,CAAC,UAAU,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;IAC7B,CAAC;CACF"}

package/build/types/criteria.d.ts

@@ -0,0 +1,24 @@
+/** Severity level assigned to a criterion (how important it is). */
+export type ModuleSeverity = "critical" | "high" | "medium" | "low";
+/** Claude model tier used for agent execution. */
+export type AgentModel = "haiku" | "sonnet" | "opus";
+/** A criterion parsed from a markdown file in the criteria directory. */
+export interface ReviewModule {
+    /** Unique identifier, e.g. "architecture/dto-enforcement". */
+    id: string;
+    /** Relative file path, e.g. "deskcheck/criteria/architecture/dto-enforcement.md". */
+    file: string;
+    /** Human-readable description from frontmatter. */
+    description: string;
+    /** How important this criterion's findings are. */
+    severity: ModuleSeverity;
+    /** File glob patterns that determine which files this criterion checks. */
+    globs: string[];
+    /** Natural language instruction for how to split files into tasks. */
+    mode: string;
+    /** Claude model tier to use for executor agents. */
+    model: AgentModel;
+    /** The detective prompt (markdown body of the criterion) given to executor agents. */
+    prompt: string;
+}
+//# sourceMappingURL=criteria.d.ts.map

package/build/types/criteria.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"criteria.d.ts","sourceRoot":"","sources":["../../src/types/criteria.ts"],"names":[],"mappings":"AAIA,oEAAoE;AACpE,MAAM,MAAM,cAAc,GAAG,UAAU,GAAG,MAAM,GAAG,QAAQ,GAAG,KAAK,CAAC;AAEpE,kDAAkD;AAClD,MAAM,MAAM,UAAU,GAAG,OAAO,GAAG,QAAQ,GAAG,MAAM,CAAC;AAErD,yEAAyE;AACzE,MAAM,WAAW,YAAY;IAC3B,8DAA8D;IAC9D,EAAE,EAAE,MAAM,CAAC;IACX,qFAAqF;IACrF,IAAI,EAAE,MAAM,CAAC;IACb,mDAAmD;IACnD,WAAW,EAAE,MAAM,CAAC;IACpB,mDAAmD;IACnD,QAAQ,EAAE,cAAc,CAAC;IACzB,2EAA2E;IAC3E,KAAK,EAAE,MAAM,EAAE,CAAC;IAChB,sEAAsE;IACtE,IAAI,EAAE,MAAM,CAAC;IACb,oDAAoD;IACpD,KAAK,EAAE,UAAU,CAAC;IAClB,sFAAsF;IACtF,MAAM,EAAE,MAAM,CAAC;CAChB"}

package/build/{core/types.js → types/criteria.js}

@@ -1,5 +1,5 @@
 // =============================================================================
-// Common Union Types
+// Criterion Types (parsed from criterion markdown files)
 // =============================================================================
 export {};
-//# sourceMappingURL=types.js.map
+//# sourceMappingURL=criteria.js.map

package/build/types/criteria.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"criteria.js","sourceRoot":"","sources":["../../src/types/criteria.ts"],"names":[],"mappings":"AAAA,gFAAgF;AAChF,yDAAyD;AACzD,gFAAgF"}

package/build/{core/types.d.ts → types/review.d.ts}

@@ -1,9 +1,6 @@
-/** Severity level assigned to a criterion (how important it is). */
-export type ModuleSeverity = "critical" | "high" | "medium" | "low";
+import type { AgentModel, ModuleSeverity } from "./criteria.js";
 /** Severity level assigned to an individual finding. */
 export type FindingSeverity = "critical" | "warning" | "info";
-/** Claude model tier used for agent execution. */
-export type AgentModel = "haiku" | "sonnet" | "opus";
 /** How the source content is provided to executors. */
 export type ContextType = "diff" | "file" | "symbol";
 /** Lifecycle status of a deskcheck plan. */
@@ -12,65 +9,6 @@ export type PlanStatus = "planning" | "ready" | "executing" | "complete";
 export type TaskStatus = "pending" | "in_progress" | "complete" | "error";
 /** Whether results cover all tasks or only a subset. */
 export type ResultsStatus = "partial" | "complete";
-/** A criterion parsed from a markdown file in the criteria directory. */
-export interface ReviewModule {
-    /** Unique identifier, e.g. "architecture/dto-enforcement". */
-    id: string;
-    /** Relative file path, e.g. "deskcheck/criteria/architecture/dto-enforcement.md". */
-    file: string;
-    /** Human-readable description from frontmatter. */
-    description: string;
-    /** How important this criterion's findings are. */
-    severity: ModuleSeverity;
-    /** File glob patterns that determine which files this criterion checks. */
-    globs: string[];
-    /** Natural language instruction for how to split files into tasks. */
-    mode: string;
-    /** Claude model tier to use for executor agents. */
-    model: AgentModel;
-    /** The detective prompt (markdown body of the criterion) given to executor agents. */
-    prompt: string;
-}
-/** MCP server configuration for agent tool access. */
-export interface McpServerConfig {
-    /** Command to start the MCP server. */
-    command: string;
-    /** Command-line arguments. */
-    args?: string[];
-    /** Environment variables passed to the server process. */
-    env?: Record<string, string>;
-}
-/** Per-role agent configuration (planner, executor, evaluator). */
-export interface AgentRoleConfig {
-    /** Override the default model for this role. */
-    model?: AgentModel;
-    /** Additional tools beyond the shared set. */
-    additional_tools?: string[];
-    /** Additional MCP servers beyond the shared set. */
-    additional_mcp_servers?: Record<string, McpServerConfig>;
-}
-/** Top-level deskcheck tool configuration from .deskcheck/config.json. */
-export interface ReviewConfig {
-    /** Directory containing criterion markdown files. */
-    modules_dir: string;
-    /** Directory for storing deskcheck run results. */
-    storage_dir: string;
-    /** Port for the web UI server. Default: 3000. */
-    serve_port: number;
-    /** Shared configuration applied to all agent roles. */
-    shared: {
-        /** Tools available to all agents. */
-        allowed_tools: string[];
-        /** MCP servers available to all agents. */
-        mcp_servers: Record<string, McpServerConfig>;
-    };
-    /** Per-role agent configuration. */
-    agents: {
-        planner: AgentRoleConfig;
-        executor: AgentRoleConfig;
-        evaluator: AgentRoleConfig;
-    };
-}
 /** What is being reviewed — a diff, file, or symbol. */
 export interface ReviewSource {
     /** The kind of content being reviewed. */
@@ -176,26 +114,54 @@ export interface TotalUsage {
     duration_api_ms: number;
     num_turns: number;
 }
-/** A single finding produced by an executor agent. */
-export interface Finding {
-    /** How severe this finding is. */
-    severity: FindingSeverity;
-    /** File path where the issue was found. */
+/** A code location referenced by an issue. */
+export interface Reference {
+    /** File path where the issue manifests. */
     file: string;
-    /** Line number where the issue occurs, if applicable. */
+    /** Semantic symbol anchor, e.g. "ClassName::method". Stable across refactors. */
+    symbol: string | null;
+    /** Line number for navigation, not the primary identifier. */
     line: number | null;
-    /** Description of the issue. */
+    /** The current code at this location (for self-contained review). */
+    code: string | null;
+    /** Suggested replacement code (optional fix). */
+    suggestedCode: string | null;
+    /** Why this reference is relevant to the issue. */
+    note: string | null;
+}
+/** A single issue produced by an executor agent. */
+export interface Issue {
+    /** Stable identity, stamped by recomputeAggregations as `${task_id}:${index}`. */
+    issue_id?: string;
+    /** How severe this issue is. */
+    severity: FindingSeverity;
+    /** Description of what's wrong. */
     description: string;
-    /** Suggested fix or improvement, if applicable. */
+    /** High-level suggested fix or improvement, if applicable. */
     suggestion: string | null;
+    /** Where this issue manifests — one or more code locations. */
+    references: Reference[];
 }
-/** A finding enriched with its source module and task for by-file grouping. */
-export interface FileFinding extends Finding {
-    /** Which criterion produced this finding. */
+/** An issue enriched with its source module and task for by-file grouping. */
+export interface FileIssue extends Issue {
+    /** Stable identity, always present on FileIssue (narrowed from optional). */
+    issue_id: string;
+    /** Which criterion produced this issue. */
     review_id: string;
-    /** Which task produced this finding. */
+    /** Which task produced this issue. */
     task_id: string;
 }
+/**
+ * Legacy Finding type for backwards compatibility with old executor output.
+ * @deprecated Use Issue instead.
+ */
+export interface Finding {
+    severity: FindingSeverity;
+    file: string;
+    line: number | null;
+    description: string;
+    suggestion: string | null;
+}
 /** Results from a single completed task. */
 export interface TaskResult {
     /** The task that produced these results. */
@@ -206,13 +172,13 @@ export interface TaskResult {
     files: string[];
     /** ISO 8601 timestamp when the task completed. */
     completed_at: string;
-    /** Findings produced by the executor agent. */
-    findings: Finding[];
+    /** Issues produced by the executor agent. */
+    issues: Issue[];
     /** Token usage and timing from the executor agent. Null for legacy results. */
     usage: TaskUsage | null;
 }
-/** Aggregated findings for a single criterion. */
-export interface ModuleFindings {
+/** Aggregated issues for a single criterion. */
+export interface ModuleIssues {
     /** Review module identifier. */
     review_id: string;
     /** Human-readable description. */
@@ -223,15 +189,15 @@ export interface ModuleFindings {
     task_count: number;
     /** Number of completed tasks. */
     completed: number;
-    /** Aggregated finding counts by severity. */
+    /** Aggregated issue counts by severity. */
     counts: {
         critical: number;
         warning: number;
         info: number;
         total: number;
     };
-    /** All findings from this module's tasks. */
-    findings: Finding[];
+    /** All issues from this module's tasks. */
+    issues: Issue[];
 }
 /** The complete deskcheck results written to results.json. */
 export interface ReviewResults {
@@ -249,7 +215,7 @@ export interface ReviewResults {
         in_progress: number;
         errored: number;
     };
-    /** Aggregated finding counts across all tasks. */
+    /** Aggregated issue counts across all tasks. */
     summary: {
         total: number;
         critical: number;
@@ -258,11 +224,11 @@ export interface ReviewResults {
     };
     /** Per-task results keyed by task_id. */
     task_results: Record<string, TaskResult>;
-    /** Findings grouped by file path. */
-    by_file: Record<string, FileFinding[]>;
-    /** Findings grouped by criterion. */
-    by_module: Record<string, ModuleFindings>;
+    /** Issues grouped by file path (an issue appears under each referenced file). */
+    by_file: Record<string, FileIssue[]>;
+    /** Issues grouped by criterion. */
+    by_module: Record<string, ModuleIssues>;
     /** Aggregated token usage across all completed tasks. */
     total_usage: TotalUsage;
 }
-//# sourceMappingURL=types.d.ts.map
+//# sourceMappingURL=review.d.ts.map

package/build/types/review.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"review.d.ts","sourceRoot":"","sources":["../../src/types/review.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,UAAU,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAMhE,wDAAwD;AACxD,MAAM,MAAM,eAAe,GAAG,UAAU,GAAG,SAAS,GAAG,MAAM,CAAC;AAE9D,uDAAuD;AACvD,MAAM,MAAM,WAAW,GAAG,MAAM,GAAG,MAAM,GAAG,QAAQ,CAAC;AAErD,4CAA4C;AAC5C,MAAM,MAAM,UAAU,GAAG,UAAU,GAAG,OAAO,GAAG,WAAW,GAAG,UAAU,CAAC;AAEzE,wDAAwD;AACxD,MAAM,MAAM,UAAU,GAAG,SAAS,GAAG,aAAa,GAAG,UAAU,GAAG,OAAO,CAAC;AAE1E,wDAAwD;AACxD,MAAM,MAAM,aAAa,GAAG,SAAS,GAAG,UAAU,CAAC;AAMnD,wDAAwD;AACxD,MAAM,WAAW,YAAY;IAC3B,0CAA0C;IAC1C,IAAI,EAAE,WAAW,CAAC;IAClB,wFAAwF;IACxF,MAAM,EAAE,MAAM,CAAC;IACf,6DAA6D;IAC7D,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED,0DAA0D;AAC1D,MAAM,WAAW,UAAU;IACzB,wEAAwE;IACxE,OAAO,EAAE,MAAM,CAAC;IAChB,qEAAqE;IACrE,SAAS,EAAE,MAAM,CAAC;IAClB,2CAA2C;IAC3C,WAAW,EAAE,MAAM,CAAC;IACpB,mCAAmC;IACnC,KAAK,EAAE,MAAM,EAAE,CAAC;IAChB,4DAA4D;IAC5D,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IACpB,gDAAgD;IAChD,KAAK,EAAE,UAAU,CAAC;IAClB,gCAAgC;IAChC,MAAM,EAAE,UAAU,CAAC;IACnB,qDAAqD;IACrD,UAAU,EAAE,MAAM,CAAC;IACnB,iDAAiD;IACjD,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,mDAAmD;IACnD,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAE5B,gFAAgF;IAChF,YAAY,EAAE,WAAW,CAAC;IAC1B,+DAA+D;IAC/D,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,wDAAwD;IACxD,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,4EAA4E;IAC5E,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;CACvB;AAED,+CAA+C;AAC/C,MAAM,WAAW,aAAa;IAC5B,gCAAgC;IAChC,SAAS,EAAE,MAAM,CAAC;IAClB,mDAAmD;IACnD,WAAW,EAAE,MAAM,CAAC;IACpB,gCAAgC;IAChC,QAAQ,EAAE,cAAc,CAAC;IACzB,0CAA0C;IAC1C,KAAK,EAAE,UAAU,CAAC;IAClB,kDAAkD;IAClD,UAAU,EAAE,MAAM,CAAC;IACnB,8CAA8C;IAC9C,aAAa,EAAE,MAAM,EAAE,CAAC;CACzB;AAED,wDAAwD;AACxD,MAAM,WAAW,UAAU;IACzB,wDAAwD;IACxD,OAAO,EAAE,MAAM,CAAC;IAChB,mEAAmE;IACnE,IAAI,EAAE,MAAM,CAAC;IACb,8BAA8B;IAC9B,MAAM,EAAE,YAAY,CAAC;IACrB,gCAAgC;IAChC,MAAM,EAAE,UAAU,CAAC;IACnB,oDAAoD;IACpD,UAAU,EAAE,MAAM,CAAC;IACnB,sDAAsD;IACtD,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,iDAAiD;IACjD,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,mDAAmD;IACnD,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAE5B,iDAAiD;IACjD,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,wDAAwD;IACxD,eAAe,EAAE,MAAM,EAAE,CAAC;IAE1B,kCAAkC;IAClC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC;IAClC,+CAA+C;IAC/C,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC;CACxC;AAMD,8DAA8D;AAC9D,MAAM,WAAW,SAAS;IACxB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,qBAAqB,EAAE,MAAM,CAAC;IAC9B,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,eAAe,EAAE,MAAM,CAAC;IACxB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;CACf;AAED,wDAAwD;AACxD,MAAM,WAAW,UAAU;IACzB,YAAY,EAAE,MAAM,CAAC;IACrB,aAAa,EAAE,MAAM,CAAC;IACtB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,qBAAqB,EAAE,MAAM,CAAC;IAC9B,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,eAAe,EAAE,MAAM,CAAC;IACxB,SAAS,EAAE,MAAM,CAAC;CACnB;AAMD,8CAA8C;AAC9C,MAAM,WAAW,SAAS;IACxB,2CAA2C;IAC3C,IAAI,EAAE,MAAM,CAAC;IACb,iFAAiF;IACjF,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,8DAA8D;IAC9D,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IACpB,qEAAqE;IACrE,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IACpB,iDAAiD;IACjD,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,mDAAmD;IACnD,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;CACrB;AAED,oDAAoD;AACpD,MAAM,WAAW,KAAK;IACpB,kFAAkF;IAClF,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,gCAAgC;IAChC,QAAQ,EAAE,eAAe,CAAC;IAC1B,mCAAmC;IACnC,WAAW,EAAE,MAAM,CAAC;IACpB,8DAA8D;IAC9D,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,+DAA+D;IAC/D,UAAU,EAAE,SAAS,EAAE,CAAC;CACzB;AAED,8EAA8E;AAC9E,MAAM,WAAW,SAAU,SAAQ,KAAK;IACtC,6EAA6E;IAC7E,QAAQ,EAAE,MAAM,CAAC;IACjB,2CAA2C;IAC3C,SAAS,EAAE,MAAM,CAAC;IAClB,sCAAsC;IACtC,OAAO,EAAE,MAAM,CAAC;CACjB;AAED;;;GAGG;AACH,MAAM,WAAW,OAAO;IACtB,QAAQ,EAAE,eAAe,CAAC;IAC1B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,GAAG,IAAI,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;CAC3B;AAED,4CAA4C;AAC5C,MAAM,WAAW,UAAU;IACzB,4CAA4C;IAC5C,OAAO,EAAE,MAAM,CAAC;IAChB,sCAAsC;IACtC,SAAS,EAAE,MAAM,CAAC;IAClB,gCAAgC;IAChC,KAAK,EAAE,MAAM,EAAE,CAAC;IAChB,kDAAkD;IAClD,YAAY,EAAE,MAAM,CAAC;IACrB,6CAA6C;IAC7C,MAAM,EAAE,KAAK,EAAE,CAAC;IAChB,+EAA+E;IAC/E,KAAK,EAAE,SAAS,GAAG,IAAI,CAAC;CACzB;AAED,gDAAgD;AAChD,MAAM,WAAW,YAAY;IAC3B,gCAAgC;IAChC,SAAS,EAAE,MAAM,CAAC;IAClB,kCAAkC;IAClC,WAAW,EAAE,MAAM,CAAC;IACpB,gCAAgC;IAChC,QAAQ,EAAE,cAAc,CAAC;IACzB,gDAAgD;IAChD,UAAU,EAAE,MAAM,CAAC;IACnB,iCAAiC;IACjC,SAAS,EAAE,MAAM,CAAC;IAClB,2CAA2C;IAC3C,MAAM,EAAE;QACN,QAAQ,EAAE,MAAM,CAAC;QACjB,OAAO,EAAE,MAAM,CAAC;QAChB,IAAI,EAAE,MAAM,CAAC;QACb,KAAK,EAAE,MAAM,CAAC;KACf,CAAC;IACF,2CAA2C;IAC3C,MAAM,EAAE,KAAK,EAAE,CAAC;CACjB;AAED,8DAA8D;AAC9D,MAAM,WAAW,aAAa;IAC5B,0CAA0C;IAC1C,OAAO,EAAE,MAAM,CAAC;IAChB,wCAAwC;IACxC,MAAM,EAAE,aAAa,CAAC;IACtB,6CAA6C;IAC7C,UAAU,EAAE,MAAM,CAAC;IAEnB,gCAAgC;IAChC,UAAU,EAAE;QACV,KAAK,EAAE,MAAM,CAAC;QACd,SAAS,EAAE,MAAM,CAAC;QAClB,OAAO,EAAE,MAAM,CAAC;QAChB,WAAW,EAAE,MAAM,CAAC;QACpB,OAAO,EAAE,MAAM,CAAC;KACjB,CAAC;IAEF,gDAAgD;IAChD,OAAO,EAAE;QACP,KAAK,EAAE,MAAM,CAAC;QACd,QAAQ,EAAE,MAAM,CAAC;QACjB,OAAO,EAAE,MAAM,CAAC;QAChB,IAAI,EAAE,MAAM,CAAC;KACd,CAAC;IAEF,yCAAyC;IACzC,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC;IACzC,iFAAiF;IACjF,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IACrC,mCAAmC;IACnC,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;IACxC,yDAAyD;IACzD,WAAW,EAAE,UAAU,CAAC;CACzB"}

package/build/types/review.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=review.js.map

package/build/types/review.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"review.js","sourceRoot":"","sources":["../../src/types/review.ts"],"names":[],"mappings":""}

package/build/types/testing.d.ts

@@ -0,0 +1,109 @@
+import type { Finding, TaskUsage } from "./review.js";
+/** Status of an individual test case. */
+export type TestCaseStatus = "pending" | "executing" | "judging" | "complete" | "error";
+/** Status of a test suite (all tests for one criterion). */
+export type TestSuiteStatus = "pending" | "running" | "complete";
+/** Status of the overall test run. */
+export type TestRunStatus = "running" | "complete";
+/** A discovered test case mapping a fixture file to expected results. */
+export interface TestCase {
+    /** Criterion identifier, e.g. "backend/controller-conventions". */
+    criterionId: string;
+    /** Test case name (subdirectory name), e.g. "missing-return-types". */
+    name: string;
+    /** Path to the criterion .md file. */
+    criterionFile: string;
+    /** Path to the fixture file (code to be reviewed). */
+    fixtureFile: string;
+    /** Path to expected.md describing what should/shouldn't be found. */
+    expectedFile: string;
+}
+/** Judge's annotation of a single executor finding. */
+export interface FindingReview {
+    /** The finding description from the executor. */
+    finding: string;
+    /** Whether the finding is correct, out of scope, or has wrong severity. */
+    verdict: "correct" | "out_of_scope" | "incorrect_severity";
+    /** Which check from the criterion this finding relates to, e.g. "#3 - Missing type hints". */
+    criterion_check: string | null;
+    /** Brief explanation of the verdict. */
+    comment: string;
+}
+/** Judge's annotation of a single expectation from expected.md. */
+export interface ExpectationReview {
+    /** The expectation text from expected.md. */
+    expectation: string;
+    /** Whether a finding matched this expectation. */
+    verdict: "found" | "missed";
+    /** Which finding matched, if any. */
+    matched_finding: string | null;
+    /** Brief explanation of the verdict. */
+    comment: string;
+}
+/** Complete judge output for a test case. */
+export interface JudgeResult {
+    /** Review of each finding the executor produced. */
+    findings_review: FindingReview[];
+    /** Review of each expectation from expected.md. */
+    expectations_review: ExpectationReview[];
+}
+/** Derived scores computed from judge verdicts. Pure math, no LLM involved. */
+export interface TestScores {
+    total_findings: number;
+    correct: number;
+    out_of_scope: number;
+    incorrect_severity: number;
+    total_expectations: number;
+    expectations_found: number;
+    expectations_missed: number;
+    /** expectations_found / total_expectations (1.0 if no expectations). */
+    recall: number;
+    /** correct / total_findings (1.0 if no findings). */
+    precision: number;
+    /** (total_findings - out_of_scope) / total_findings (1.0 if no findings). */
+    scope_compliance: number;
+}
+/** Result of a single test case, persisted in the test run JSON. */
+export interface TestCaseResult {
+    /** Current lifecycle status. */
+    status: TestCaseStatus;
+    /** Path to the fixture file that was reviewed. */
+    fixture_file: string;
+    /** Path to the expected.md file. */
+    expected_file: string;
+    /** Findings produced by the executor agent. */
+    findings: Finding[] | null;
+    /** Judge's evaluation of findings vs expectations. */
+    judge: JudgeResult | null;
+    /** Derived scores from judge verdicts. */
+    scores: TestScores | null;
+    /** Error message if the test case failed. */
+    error: string | null;
+    /** Token usage from the executor agent run. */
+    executor_usage: TaskUsage | null;
+    /** Token usage from the judge agent run. */
+    judge_usage: TaskUsage | null;
+}
+/** A test suite groups all test cases for one criterion. */
+export interface TestSuite {
+    /** Current lifecycle status. */
+    status: TestSuiteStatus;
+    /** Path to the criterion .md file. */
+    criterion_file: string;
+    /** Test case results keyed by test name. */
+    tests: Record<string, TestCaseResult>;
+}
+/** The complete test run persisted to disk as results.json. */
+export interface TestRun {
+    /** Unique run identifier (timestamped directory name). */
+    run_id: string;
+    /** Current lifecycle status. */
+    status: TestRunStatus;
+    /** ISO 8601 timestamp when the run started. */
+    started_at: string;
+    /** ISO 8601 timestamp when the run completed, or null if still running. */
+    completed_at: string | null;
+    /** Test suites keyed by criterionId. */
+    suites: Record<string, TestSuite>;
+}
+//# sourceMappingURL=testing.d.ts.map

package/build/types/testing.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"testing.d.ts","sourceRoot":"","sources":["../../src/types/testing.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAMtD,yCAAyC;AACzC,MAAM,MAAM,cAAc,GAAG,SAAS,GAAG,WAAW,GAAG,SAAS,GAAG,UAAU,GAAG,OAAO,CAAC;AAExF,4DAA4D;AAC5D,MAAM,MAAM,eAAe,GAAG,SAAS,GAAG,SAAS,GAAG,UAAU,CAAC;AAEjE,sCAAsC;AACtC,MAAM,MAAM,aAAa,GAAG,SAAS,GAAG,UAAU,CAAC;AAMnD,yEAAyE;AACzE,MAAM,WAAW,QAAQ;IACvB,mEAAmE;IACnE,WAAW,EAAE,MAAM,CAAC;IACpB,uEAAuE;IACvE,IAAI,EAAE,MAAM,CAAC;IACb,sCAAsC;IACtC,aAAa,EAAE,MAAM,CAAC;IACtB,sDAAsD;IACtD,WAAW,EAAE,MAAM,CAAC;IACpB,qEAAqE;IACrE,YAAY,EAAE,MAAM,CAAC;CACtB;AAMD,uDAAuD;AACvD,MAAM,WAAW,aAAa;IAC5B,iDAAiD;IACjD,OAAO,EAAE,MAAM,CAAC;IAChB,2EAA2E;IAC3E,OAAO,EAAE,SAAS,GAAG,cAAc,GAAG,oBAAoB,CAAC;IAC3D,8FAA8F;IAC9F,eAAe,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,wCAAwC;IACxC,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,mEAAmE;AACnE,MAAM,WAAW,iBAAiB;IAChC,6CAA6C;IAC7C,WAAW,EAAE,MAAM,CAAC;IACpB,kDAAkD;IAClD,OAAO,EAAE,OAAO,GAAG,QAAQ,CAAC;IAC5B,qCAAqC;IACrC,eAAe,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,wCAAwC;IACxC,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,6CAA6C;AAC7C,MAAM,WAAW,WAAW;IAC1B,oDAAoD;IACpD,eAAe,EAAE,aAAa,EAAE,CAAC;IACjC,mDAAmD;IACnD,mBAAmB,EAAE,iBAAiB,EAAE,CAAC;CAC1C;AAMD,+EAA+E;AAC/E,MAAM,WAAW,UAAU;IACzB,cAAc,EAAE,MAAM,CAAC;IACvB,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;IACrB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,kBAAkB,EAAE,MAAM,CAAC;IAC3B,kBAAkB,EAAE,MAAM,CAAC;IAC3B,mBAAmB,EAAE,MAAM,CAAC;IAC5B,wEAAwE;IACxE,MAAM,EAAE,MAAM,CAAC;IACf,qDAAqD;IACrD,SAAS,EAAE,MAAM,CAAC;IAClB,6EAA6E;IAC7E,gBAAgB,EAAE,MAAM,CAAC;CAC1B;AAMD,oEAAoE;AACpE,MAAM,WAAW,cAAc;IAC7B,gCAAgC;IAChC,MAAM,EAAE,cAAc,CAAC;IACvB,kDAAkD;IAClD,YAAY,EAAE,MAAM,CAAC;IACrB,oCAAoC;IACpC,aAAa,EAAE,MAAM,CAAC;IACtB,+CAA+C;IAC/C,QAAQ,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC;IAC3B,sDAAsD;IACtD,KAAK,EAAE,WAAW,GAAG,IAAI,CAAC;IAC1B,0CAA0C;IAC1C,MAAM,EAAE,UAAU,GAAG,IAAI,CAAC;IAC1B,6CAA6C;IAC7C,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,+CAA+C;IAC/C,cAAc,EAAE,SAAS,GAAG,IAAI,CAAC;IACjC,4CAA4C;IAC5C,WAAW,EAAE,SAAS,GAAG,IAAI,CAAC;CAC/B;AAMD,4DAA4D;AAC5D,MAAM,WAAW,SAAS;IACxB,gCAAgC;IAChC,MAAM,EAAE,eAAe,CAAC;IACxB,sCAAsC;IACtC,cAAc,EAAE,MAAM,CAAC;IACvB,4CAA4C;IAC5C,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;CACvC;AAED,+DAA+D;AAC/D,MAAM,WAAW,OAAO;IACtB,0DAA0D;IAC1D,MAAM,EAAE,MAAM,CAAC;IACf,gCAAgC;IAChC,MAAM,EAAE,aAAa,CAAC;IACtB,+CAA+C;IAC/C,UAAU,EAAE,MAAM,CAAC;IACnB,2EAA2E;IAC3E,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,wCAAwC;IACxC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;CACnC"}

package/build/types/testing.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=testing.js.map

package/build/types/testing.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"testing.js","sourceRoot":"","sources":["../../src/types/testing.ts"],"names":[],"mappings":""}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@prowi/deskcheck",
-  "version": "0.1.0",
+  "version": "0.3.0",
   "description": "Modular code review tool powered by Claude Agent SDK — define review criteria as markdown, run them against your codebase, get structured findings with a web dashboard.",
   "license": "MIT",
   "repository": {