@agentv/core 4.20.0 → 4.21.0-next.1

package/dist/index.d.cts

@@ -4082,7 +4082,7 @@ declare function createDraftResultsPr(params: {
 /**
  * The default config directory (~/.agentv). Always resolves to the user's home
  * directory regardless of AGENTV_HOME. Used for lightweight, machine-local files
- * like version-check.json, last-config.json, and projects.yaml.
+ * like version-check.json, last-config.json, and benchmarks.yaml.
  */
 declare function getAgentvConfigDir(): string;
 /**
@@ -4100,18 +4100,31 @@ declare function getWorkspacePoolRoot(): string;
  * Benchmark registry for AgentV Studio multi-benchmark support.
  *
  * A Benchmark = any directory containing a `.agentv/` folder.
- * The registry lives at `~/.agentv/projects.yaml` and tracks registered benchmarks.
+ * The registry lives at `~/.agentv/benchmarks.yaml` and is the single source of
+ * truth for which benchmarks Studio shows. Studio re-reads the file on every
+ * `/api/benchmarks` request, so edits (direct, via POST /api/benchmarks, via
+ * the CLI's --add/--remove, or via a Kubernetes ConfigMap mount) are reflected
+ * without restarting `agentv serve`.
  *
- * YAML format:
+ * YAML format (all keys snake_case per AGENTS.md §"Wire Format Convention"):
  *   benchmarks:
  *     - id: my-app
  *       name: My App
  *       path: /home/user/projects/my-app
- *       addedAt: "2026-03-20T10:00:00Z"
- *       lastOpenedAt: "2026-03-30T14:00:00Z"
+ *       added_at: "2026-03-20T10:00:00Z"
+ *       last_opened_at: "2026-03-30T14:00:00Z"
  *
- * To extend: use loadBenchmarkRegistry() / saveBenchmarkRegistry() for CRUD,
- * discoverBenchmarks() to scan a directory tree for `.agentv/` directories.
+ * Concurrency: the registry assumes a single writer. All mutating calls
+ * (add/remove/touchBenchmark) do read-modify-write on benchmarks.yaml
+ * without a lock. Studio's HTTP handlers are serialized by Node's
+ * single-threaded event loop, which satisfies the 24/7 deployment case.
+ * Run only one `agentv` process against a given home at a time.
+ *
+ * To extend:
+ *   - CRUD: loadBenchmarkRegistry() / saveBenchmarkRegistry() + the
+ *     add/remove/touch helpers.
+ *   - discoverBenchmarks() is a one-shot filesystem utility for bulk
+ *     registration; it does not run in the request path.
  */
 interface BenchmarkEntry {
     id: string;
@@ -4151,7 +4164,9 @@ declare function getBenchmark(benchmarkId: string): BenchmarkEntry | undefined;
 declare function touchBenchmark(benchmarkId: string): void;
 /**
  * Scan a directory tree (up to maxDepth levels) for directories containing `.agentv/`.
- * Returns absolute paths of discovered benchmark directories.
+ * Returns absolute paths of discovered benchmark directories, sorted for
+ * deterministic iteration. This is a one-shot helper for bulk registration;
+ * Studio does not scan at request time.
  */
 declare function discoverBenchmarks(rootDir: string, maxDepth?: number): string[];
 

package/dist/index.d.ts

@@ -4082,7 +4082,7 @@ declare function createDraftResultsPr(params: {
 /**
  * The default config directory (~/.agentv). Always resolves to the user's home
  * directory regardless of AGENTV_HOME. Used for lightweight, machine-local files
- * like version-check.json, last-config.json, and projects.yaml.
+ * like version-check.json, last-config.json, and benchmarks.yaml.
  */
 declare function getAgentvConfigDir(): string;
 /**
@@ -4100,18 +4100,31 @@ declare function getWorkspacePoolRoot(): string;
  * Benchmark registry for AgentV Studio multi-benchmark support.
  *
  * A Benchmark = any directory containing a `.agentv/` folder.
- * The registry lives at `~/.agentv/projects.yaml` and tracks registered benchmarks.
+ * The registry lives at `~/.agentv/benchmarks.yaml` and is the single source of
+ * truth for which benchmarks Studio shows. Studio re-reads the file on every
+ * `/api/benchmarks` request, so edits (direct, via POST /api/benchmarks, via
+ * the CLI's --add/--remove, or via a Kubernetes ConfigMap mount) are reflected
+ * without restarting `agentv serve`.
  *
- * YAML format:
+ * YAML format (all keys snake_case per AGENTS.md §"Wire Format Convention"):
  *   benchmarks:
  *     - id: my-app
  *       name: My App
  *       path: /home/user/projects/my-app
- *       addedAt: "2026-03-20T10:00:00Z"
- *       lastOpenedAt: "2026-03-30T14:00:00Z"
+ *       added_at: "2026-03-20T10:00:00Z"
+ *       last_opened_at: "2026-03-30T14:00:00Z"
  *
- * To extend: use loadBenchmarkRegistry() / saveBenchmarkRegistry() for CRUD,
- * discoverBenchmarks() to scan a directory tree for `.agentv/` directories.
+ * Concurrency: the registry assumes a single writer. All mutating calls
+ * (add/remove/touchBenchmark) do read-modify-write on benchmarks.yaml
+ * without a lock. Studio's HTTP handlers are serialized by Node's
+ * single-threaded event loop, which satisfies the 24/7 deployment case.
+ * Run only one `agentv` process against a given home at a time.
+ *
+ * To extend:
+ *   - CRUD: loadBenchmarkRegistry() / saveBenchmarkRegistry() + the
+ *     add/remove/touch helpers.
+ *   - discoverBenchmarks() is a one-shot filesystem utility for bulk
+ *     registration; it does not run in the request path.
  */
 interface BenchmarkEntry {
     id: string;
@@ -4151,7 +4164,9 @@ declare function getBenchmark(benchmarkId: string): BenchmarkEntry | undefined;
 declare function touchBenchmark(benchmarkId: string): void;
 /**
  * Scan a directory tree (up to maxDepth levels) for directories containing `.agentv/`.
- * Returns absolute paths of discovered benchmark directories.
+ * Returns absolute paths of discovered benchmark directories, sorted for
+ * deterministic iteration. This is a one-shot helper for bulk registration;
+ * Studio does not scan at request time.
  */
 declare function discoverBenchmarks(rootDir: string, maxDepth?: number): string[];
 

package/dist/index.js

@@ -128,7 +128,7 @@ import {
   toCamelCaseDeep,
   toSnakeCaseDeep,
   tokensPerTool
-} from "./chunk-ELF6SQAK.js";
+} from "./chunk-WCW3V6QJ.js";
 import {
   COMMON_TARGET_SETTINGS,
   TEST_MESSAGE_ROLES,
@@ -152,7 +152,7 @@ import {
   resolveDelegatedTargetDefinition,
   resolveFileReference,
   resolveTargetDefinition
-} from "./chunk-24ND5HZC.js";
+} from "./chunk-LKX4QW3G.js";
 import "./chunk-3WGHC7LC.js";
 import "./chunk-PRNXHNLF.js";
 import {
@@ -992,44 +992,48 @@ async function createDraftResultsPr(params) {
 }
 
 // src/benchmarks.ts
-import {
-  copyFileSync,
-  existsSync as existsSync2,
-  mkdirSync as mkdirSync2,
-  readFileSync as readFileSync3,
-  readdirSync,
-  statSync,
-  writeFileSync as writeFileSync2
-} from "node:fs";
+import { existsSync as existsSync2, mkdirSync as mkdirSync2, readFileSync as readFileSync3, readdirSync, statSync, writeFileSync as writeFileSync2 } from "node:fs";
 import path5 from "node:path";
 import { parse as parseYaml, stringify as stringifyYaml } from "yaml";
 function getBenchmarksRegistryPath() {
-  return path5.join(getAgentvConfigDir(), "projects.yaml");
+  return path5.join(getAgentvConfigDir(), "benchmarks.yaml");
 }
-function migrateProjectsYaml(targetPath) {
-  const dataHome = getAgentvHome();
-  const configDir = getAgentvConfigDir();
-  if (dataHome === configDir) return;
-  const legacyPath = path5.join(dataHome, "projects.yaml");
-  if (!existsSync2(legacyPath)) return;
-  mkdirSync2(path5.dirname(targetPath), { recursive: true });
-  copyFileSync(legacyPath, targetPath);
+function fromYaml(raw) {
+  if (!raw || typeof raw !== "object") return null;
+  const e = raw;
+  if (typeof e.id !== "string" || typeof e.name !== "string" || typeof e.path !== "string") {
+    return null;
+  }
+  return {
+    id: e.id,
+    name: e.name,
+    path: e.path,
+    addedAt: typeof e.added_at === "string" ? e.added_at : "",
+    lastOpenedAt: typeof e.last_opened_at === "string" ? e.last_opened_at : ""
+  };
+}
+function toYaml(entry) {
+  return {
+    id: entry.id,
+    name: entry.name,
+    path: entry.path,
+    added_at: entry.addedAt,
+    last_opened_at: entry.lastOpenedAt
+  };
 }
 function loadBenchmarkRegistry() {
   const registryPath = getBenchmarksRegistryPath();
-  if (!existsSync2(registryPath)) {
-    migrateProjectsYaml(registryPath);
-  }
   if (!existsSync2(registryPath)) {
     return { benchmarks: [] };
   }
   try {
     const raw = readFileSync3(registryPath, "utf-8");
     const parsed = parseYaml(raw);
-    if (!parsed || !Array.isArray(parsed.benchmarks)) {
+    if (!parsed || typeof parsed !== "object") {
       return { benchmarks: [] };
     }
-    return { benchmarks: parsed.benchmarks };
+    const benchmarks = Array.isArray(parsed.benchmarks) ? parsed.benchmarks.map(fromYaml).filter((e) => e !== null) : [];
+    return { benchmarks };
   } catch {
     return { benchmarks: [] };
   }
@@ -1040,7 +1044,8 @@ function saveBenchmarkRegistry(registry) {
   if (!existsSync2(dir)) {
     mkdirSync2(dir, { recursive: true });
   }
-  writeFileSync2(registryPath, stringifyYaml({ benchmarks: registry.benchmarks }), "utf-8");
+  const payload = { benchmarks: registry.benchmarks.map(toYaml) };
+  writeFileSync2(registryPath, stringifyYaml(payload), "utf-8");
 }
 function deriveBenchmarkId(dirPath, existingIds) {
   const base = path5.basename(dirPath).toLowerCase().replace(/[^a-z0-9-]/g, "-").replace(/-+/g, "-").replace(/^-|-$/g, "");
@@ -1123,7 +1128,7 @@ function discoverBenchmarks(rootDir, maxDepth = 2) {
     }
   }
   scan(absRoot, 0);
-  return results;
+  return results.sort();
 }
 
 // src/evaluation/baseline.ts