@gaunt-sloth/batch 2.0.0-alpha.19 → 2.0.0-alpha.21

package/dist/toolChecks.js

@@ -0,0 +1,38 @@
+import { toolNameMatchesPattern } from '@gaunt-sloth/core/utils/toolMatching.js';
+/**
+ * BATCH-10 tool-trace assertions — grade a case against the tool *names* it actually invoked
+ * (`cellResult.tools`), rather than grepping the answer text. This "kills the false-positive class
+ * structurally": an MCP-only case can prove the server was called instead of a substring that a
+ * hallucinated answer could also contain.
+ *
+ * Kept separate from `runDeterministicChecks` (which reads the *answer*) on purpose — these read the
+ * tool trace — so neither signature is muddied. Patterns reuse GS2-61's `toolNameMatchesPattern`
+ * (`@gaunt-sloth/core`), the same glob/exact matcher `allowedTools` uses, so `mcp__unimarket__*`
+ * behaves identically here.
+ *
+ * - `mustCall` — for **each** pattern, at least one called tool must match it, else
+ *   `did not call "<pattern>"`.
+ * - `mustNotCall` — **no** called tool may match any forbidden pattern; each offending tool is
+ *   reported once as `called forbidden tool "<tool>" (matched "<pattern>")`.
+ *
+ * BATCH-12: grades one {@link EvalExpectation} block's tool-trace assertions (a flat case's single
+ * block or a matrix case's identity-scoped block) — same field names, same behavior as before.
+ */
+export function runToolCallChecks(tools, expectation) {
+    const failures = [];
+    for (const pattern of expectation.mustCall) {
+        if (!tools.some((tool) => toolNameMatchesPattern(tool, pattern))) {
+            failures.push(`did not call "${pattern}"`);
+        }
+    }
+    // De-duplicate: a tool called N times (traces realistically repeat names) is one violation, not
+    // N identical failure lines. `Set` preserves first-seen order.
+    for (const tool of new Set(tools)) {
+        const matched = expectation.mustNotCall.find((pattern) => toolNameMatchesPattern(tool, pattern));
+        if (matched !== undefined) {
+            failures.push(`called forbidden tool "${tool}" (matched "${matched}")`);
+        }
+    }
+    return failures;
+}
+//# sourceMappingURL=toolChecks.js.map

package/dist/toolChecks.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"toolChecks.js","sourceRoot":"","sources":["../src/toolChecks.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,sBAAsB,EAAE,MAAM,yCAAyC,CAAC;AAIjF;;;;;;;;;;;;;;;;;;GAkBG;AACH,MAAM,UAAU,iBAAiB,CAC/B,KAAe,EACf,WAA8D;IAE9D,MAAM,QAAQ,GAAa,EAAE,CAAC;IAE9B,KAAK,MAAM,OAAO,IAAI,WAAW,CAAC,QAAQ,EAAE,CAAC;QAC3C,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,sBAAsB,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,EAAE,CAAC;YACjE,QAAQ,CAAC,IAAI,CAAC,iBAAiB,OAAO,GAAG,CAAC,CAAC;QAC7C,CAAC;IACH,CAAC;IAED,gGAAgG;IAChG,+DAA+D;IAC/D,KAAK,MAAM,IAAI,IAAI,IAAI,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;QAClC,MAAM,OAAO,GAAG,WAAW,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE,CACvD,sBAAsB,CAAC,IAAI,EAAE,OAAO,CAAC,CACtC,CAAC;QACF,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;YAC1B,QAAQ,CAAC,IAAI,CAAC,0BAA0B,IAAI,eAAe,OAAO,IAAI,CAAC,CAAC;QAC1E,CAAC;IACH,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gaunt-sloth/batch",
-  "version": "2.0.0-alpha.19",
+  "version": "2.0.0-alpha.21",
   "description": "Batch matrix runtime for Gaunt Sloth (BATCH-1): run a prompt-executable over a matrix of models and/or content-bound inputs",
   "license": "MIT",
   "author": "Andrew Kondratev",
@@ -23,6 +23,9 @@
   "type": "module",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
+  "bin": {
+    "gth-batch": "dist/bin.js"
+  },
   "exports": {
     ".": "./dist/index.js",
     "./*.js": "./dist/*.js"
@@ -34,8 +37,8 @@
     "@langchain/core": "^1.2.3",
     "yaml": "^2.9.0",
     "zod": "^4.4.3",
-    "@gaunt-sloth/agent": "2.0.0-alpha.19",
-    "@gaunt-sloth/core": "2.0.0-alpha.19"
+    "@gaunt-sloth/agent": "2.0.0-alpha.21",
+    "@gaunt-sloth/core": "2.0.0-alpha.21"
   },
   "files": [
     "./dist/*"