@agentgrader/scorer-static 0.1.0

package/dist/index.d.ts

@@ -0,0 +1,28 @@
+import { Scorer, AgentResult, ScorerResult } from '@agentgrader/core';
+
+/**
+ * deterministic, additive code-quality scorer: diff size, files touched,
+ * todo/fixme markers introduced, and biome lint violations on the changed
+ * files. never blocks a run (`passed` is always `true`) - it only annotates
+ * `metrics["static-quality"]` with `quality` data for reporting/optimizer
+ * use.
+ *
+ * no llm calls. lint runs via `npx @biomejs/biome` inside the sandbox; if
+ * that's unavailable (no network, image without npm), `linterviolations`
+ * degrades to `0` rather than failing the run.
+ */
+declare class StaticQualityScorer implements Scorer {
+    readonly name = "static-quality";
+    score(input: {
+        result: AgentResult;
+        sandbox: {
+            exec(cmd: string): Promise<{
+                stdout: string;
+                stderr: string;
+                exitCode: number;
+            }>;
+        };
+    }): Promise<ScorerResult>;
+}
+
+export { StaticQualityScorer };

package/dist/index.js

@@ -0,0 +1,52 @@
+import { parseDiffStats } from '@agentgrader/core';
+
+// src/index.ts
+var TODO_PATTERN = /^\+.*\b(TODO|FIXME|HACK|XXX)\b/gm;
+var StaticQualityScorer = class {
+  name = "static-quality";
+  async score(input) {
+    const diff = input.result.finalDiff ?? "";
+    const stats = parseDiffStats(diff);
+    const diffLines = stats.linesChanged;
+    const filesModified = stats.filesChanged.length;
+    const todosIntroduced = (diff.match(TODO_PATTERN) ?? []).length;
+    let linterViolations = 0;
+    if (stats.filesChanged.length > 0) {
+      try {
+        const fileArgs = stats.filesChanged.map((f) => `"${f}"`).join(" ");
+        const res = await input.sandbox.exec(
+          `npx --yes @biomejs/biome@2 check ${fileArgs} --reporter=json 2>/dev/null || echo '{}'`
+        );
+        linterViolations = parseBiomeViolationCount(res.stdout);
+      } catch {
+      }
+    }
+    return {
+      passed: true,
+      detail: `diff:${diffLines}L files:${filesModified} todos:${todosIntroduced} lint:${linterViolations}`,
+      quality: {
+        diffLines,
+        filesModified,
+        todosIntroduced,
+        linterViolations
+      }
+    };
+  }
+};
+function parseBiomeViolationCount(output) {
+  try {
+    const parsed = JSON.parse(output);
+    const errors = parsed?.summary?.errors;
+    const warnings = parsed?.summary?.warnings;
+    if (typeof errors === "number" || typeof warnings === "number") {
+      return (errors ?? 0) + (warnings ?? 0);
+    }
+    if (Array.isArray(parsed?.diagnostics)) {
+      return parsed.diagnostics.length;
+    }
+  } catch {
+  }
+  return 0;
+}
+
+export { StaticQualityScorer };

package/package.json

@@ -0,0 +1,32 @@
+{
+  "name": "@agentgrader/scorer-static",
+  "version": "0.1.0",
+  "description": "Deterministic, non-blocking code-quality scorer (diff size, TODOs, lint) for the Agentgrader framework",
+  "license": "MIT",
+  "type": "module",
+  "main": "./dist/index.js",
+  "module": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "import": "./dist/index.js",
+      "types": "./dist/index.d.ts"
+    }
+  },
+  "files": [
+    "dist"
+  ],
+  "scripts": {
+    "build": "tsup src/index.ts --format esm --dts --clean --treeshake",
+    "build:watch": "tsup src/index.ts --format esm --dts --watch"
+  },
+  "dependencies": {
+    "@agentgrader/core": "^1.1.0"
+  },
+  "devDependencies": {
+    "tsup": "^8.5.1"
+  },
+  "peerDependencies": {
+    "@agentgrader/core": "^1.1.0"
+  }
+}