Improve benchmark case details and score chart scaling (#146)

Co-authored-by: Yaowei Zheng <hiyouga@buaa.edu.cn>
2026-07-31 20:42:32 +08:00
parent 5308002d72
commit 740f101242
19 changed files with 519 additions and 212 deletions
@@ -1292,6 +1292,8 @@ export interface BenchmarksResponse {
  benchmarks: BenchmarkSummary[];
 }

+export type CaseMaterial = "statement" | "rubric";
+
 /** Public Benchmark Case metadata. Rubric and Gold content are never included. */
 export interface BenchmarkCaseSummary {
  id: string;
@@ -4,16 +4,72 @@
 *   GET /api/projects/:p/agents/:a/benchmarks/:benchmarkId/cases
 *   GET /api/projects/:p/agents/:a/benchmarks/:benchmarkId/cases/:caseId/files
 *   GET /api/projects/:p/agents/:a/benchmarks/:benchmarkId/cases/:caseId/files/content
+ *   GET /api/projects/:p/agents/:a/benchmarks/:benchmarkId/cases/:caseId/rubric/files
+ *   GET /api/projects/:p/agents/:a/benchmarks/:benchmarkId/cases/:caseId/rubric/files/content
 * Returns the Agent's Benchmark list (title/description from benchmark_config.toml)
 * along with the evaluations[] from scoreboard.yaml.
 */
-import { Hono } from "hono";
+import { Hono, type Context } from "hono";
 import type { AppEnv } from "../../auth/middleware.js";
 import type { AppDeps } from "../../app.js";
+import type { CaseMaterial } from "../../api/types.js";
 import { requireValidId } from "../validate.js";

 const TEXT_PREVIEW_BYTES = 256 * 1024;

+function listCaseFiles(deps: AppDeps, material: CaseMaterial) {
+  return async (c: Context<AppEnv>) => {
+    const projectId = requireValidId(c, "projectId");
+    const agentId = requireValidId(c, "agentId");
+    const benchmarkId = requireValidId(c, "benchmarkId");
+    const caseId = requireValidId(c, "caseId");
+    deps.projectService.requireProjectAccess(c.var.user.userId, projectId);
+    await deps.agentConfigService.requireExists(projectId, agentId);
+    return c.json(
+      await deps.benchmarks.listCaseFiles(
+        projectId,
+        agentId,
+        benchmarkId,
+        caseId,
+        c.req.query("path") ?? "",
+        material,
+      ),
+    );
+  };
+}
+
+function readCaseFile(deps: AppDeps, material: CaseMaterial) {
+  return async (c: Context<AppEnv>) => {
+    const projectId = requireValidId(c, "projectId");
+    const agentId = requireValidId(c, "agentId");
+    const benchmarkId = requireValidId(c, "benchmarkId");
+    const caseId = requireValidId(c, "caseId");
+    deps.projectService.requireProjectAccess(c.var.user.userId, projectId);
+    await deps.agentConfigService.requireExists(projectId, agentId);
+    const download = c.req.query("download") === "1";
+    const boundedPreview = !download && c.req.query("preview") === "1";
+    const { data, fileName, contentType, scriptable, truncated } =
+      await deps.benchmarks.readCaseFile(
+        projectId,
+        agentId,
+        benchmarkId,
+        caseId,
+        c.req.query("path") ?? "",
+        material,
+        boundedPreview ? { maxBytes: TEXT_PREVIEW_BYTES } : undefined,
+      );
+    return new Response(new Uint8Array(data), {
+      status: 200,
+      headers: {
+        "Content-Type": !download && scriptable ? "text/plain; charset=utf-8" : contentType,
+        "Content-Disposition": `${download ? "attachment" : "inline"}; filename*=UTF-8''${encodeURIComponent(fileName)}`,
+        "X-Content-Type-Options": "nosniff",
+        ...(truncated ? { "X-Content-Truncated": "1" } : {}),
+      },
+    });
+  };
+}
+
 export function benchmarksRoutes(deps: AppDeps): Hono<AppEnv> {
  const app = new Hono<AppEnv>();

@@ -34,52 +90,10 @@ export function benchmarksRoutes(deps: AppDeps): Hono<AppEnv> {
    return c.json(await deps.benchmarks.listCases(projectId, agentId, benchmarkId));
  });

-  app.get("/:benchmarkId/cases/:caseId/files", async (c) => {
-    const projectId = requireValidId(c, "projectId");
-    const agentId = requireValidId(c, "agentId");
-    const benchmarkId = requireValidId(c, "benchmarkId");
-    const caseId = requireValidId(c, "caseId");
-    deps.projectService.requireProjectAccess(c.var.user.userId, projectId);
-    await deps.agentConfigService.requireExists(projectId, agentId);
-    return c.json(
-      await deps.benchmarks.listCaseFiles(
-        projectId,
-        agentId,
-        benchmarkId,
-        caseId,
-        c.req.query("path") ?? "",
-      ),
-    );
-  });
-
-  app.get("/:benchmarkId/cases/:caseId/files/content", async (c) => {
-    const projectId = requireValidId(c, "projectId");
-    const agentId = requireValidId(c, "agentId");
-    const benchmarkId = requireValidId(c, "benchmarkId");
-    const caseId = requireValidId(c, "caseId");
-    deps.projectService.requireProjectAccess(c.var.user.userId, projectId);
-    await deps.agentConfigService.requireExists(projectId, agentId);
-    const download = c.req.query("download") === "1";
-    const boundedPreview = !download && c.req.query("preview") === "1";
-    const { data, fileName, contentType, scriptable, truncated } =
-      await deps.benchmarks.readCaseFile(
-        projectId,
-        agentId,
-        benchmarkId,
-        caseId,
-        c.req.query("path") ?? "",
-        boundedPreview ? { maxBytes: TEXT_PREVIEW_BYTES } : undefined,
-      );
-    return new Response(new Uint8Array(data), {
-      status: 200,
-      headers: {
-        "Content-Type": !download && scriptable ? "text/plain; charset=utf-8" : contentType,
-        "Content-Disposition": `${download ? "attachment" : "inline"}; filename*=UTF-8''${encodeURIComponent(fileName)}`,
-        "X-Content-Type-Options": "nosniff",
-        ...(truncated ? { "X-Content-Truncated": "1" } : {}),
-      },
-    });
-  });
+  app.get("/:benchmarkId/cases/:caseId/files", listCaseFiles(deps, "statement"));
+  app.get("/:benchmarkId/cases/:caseId/files/content", readCaseFile(deps, "statement"));
+  app.get("/:benchmarkId/cases/:caseId/rubric/files", listCaseFiles(deps, "rubric"));
+  app.get("/:benchmarkId/cases/:caseId/rubric/files/content", readCaseFile(deps, "rubric"));

  return app;
 }
@@ -25,6 +25,7 @@ import type {
  BenchmarkRunScore,
  BenchmarkSummary,
  BenchmarksResponse,
+  CaseMaterial,
  WorkspaceFilesResponse,
 } from "../api/types.js";
 import type {
@@ -235,7 +236,13 @@ export class BenchmarkService {
      .sort((a, b) => a.name.localeCompare(b.name))) {
      const fallback: BenchmarkCaseSummary = { id: entry.name, title: entry.name };
      try {
-        const statementDir = await this.statementRoot(projectId, agentId, benchmarkId, entry.name);
+        const statementDir = await this.caseMaterialRoot(
+          projectId,
+          agentId,
+          benchmarkId,
+          entry.name,
+          "statement",
+        );
        const realReadme = await fs.realpath(path.join(statementDir, "README.md"));
        if (!isWithin(statementDir, realReadme)) throw new Error("README escapes Statement");
        cases.push({
@@ -255,9 +262,16 @@ export class BenchmarkService {
    benchmarkId: string,
    caseId: string,
    rel: string,
+    material: CaseMaterial,
  ): Promise<WorkspaceFilesResponse> {
-    const statementDir = await this.statementRoot(projectId, agentId, benchmarkId, caseId);
-    return this.workspaceFiles.list(statementDir, rel);
+    const materialRoot = await this.caseMaterialRoot(
+      projectId,
+      agentId,
+      benchmarkId,
+      caseId,
+      material,
+    );
+    return this.workspaceFiles.list(materialRoot, rel);
  }

  async readCaseFile(
@@ -266,37 +280,45 @@ export class BenchmarkService {
    benchmarkId: string,
    caseId: string,
    rel: string,
+    material: CaseMaterial,
    options?: WorkspaceFileReadOptions,
  ): Promise<WorkspaceFileContent> {
-    const statementDir = await this.statementRoot(projectId, agentId, benchmarkId, caseId);
-    return this.workspaceFiles.read(statementDir, rel, options);
+    const materialRoot = await this.caseMaterialRoot(
+      projectId,
+      agentId,
+      benchmarkId,
+      caseId,
+      material,
+    );
+    return this.workspaceFiles.read(materialRoot, rel, options);
  }

-  private async statementRoot(
+  private async caseMaterialRoot(
    projectId: string,
    agentId: string,
    benchmarkId: string,
    caseId: string,
+    material: CaseMaterial,
  ): Promise<string> {
    const benchDir = path.join(benchmarksDir(this.root, projectId, agentId), benchmarkId);
    const caseDir = path.join(benchDir, caseId);
-    const statementDir = path.join(caseDir, "statement");
+    const materialRoot = path.join(caseDir, material);
    try {
-      const [realBenchDir, realCaseDir, realStatementDir] = await Promise.all([
+      const [realBenchDir, realCaseDir, realMaterialRoot] = await Promise.all([
        fs.realpath(benchDir),
        fs.realpath(caseDir),
-        fs.realpath(statementDir),
+        fs.realpath(materialRoot),
      ]);
      if (
        !isWithin(realBenchDir, realCaseDir) ||
-        path.dirname(realStatementDir) !== realCaseDir ||
-        path.basename(realStatementDir) !== "statement"
+        path.dirname(realMaterialRoot) !== realCaseDir ||
+        path.basename(realMaterialRoot) !== material
      ) {
-        throw new Error("Statement path is not canonical");
+        throw new Error("Case material is not canonical");
      }
-      return realStatementDir;
+      return realMaterialRoot;
    } catch {
-      throw new HttpError(404, "not_found", "Public Case Statement does not exist.");
+      throw new HttpError(404, "not_found", `Case ${material} does not exist.`);
    }
  }