@gaunt-sloth/batch 2.0.0-beta.10 → 2.0.0-beta.12

package/dist/types.d.ts

@@ -7,6 +7,7 @@
  * `packages/app/src/commands/batchCommand.ts`, which is what keeps {@link RunCellFn} injectable —
  * unit tests here fake it, the CLI wires the real one.
  */
+import type { AdvertisedToolInventory } from '#src/toolCoverage.js';
 /** One row/record parsed from `--over <path.csv|path.jsonl>`. Values are always strings for csv. */
 export type MatrixRow = Record<string, string>;
 /**
@@ -46,6 +47,17 @@ export interface ToolResultRecord {
     isError: boolean;
     /** The result payload as text (non-string payloads JSON-stringified), size-capped at capture. */
     content?: string;
+    /**
+     * BATCH-43 — an errored MCP tool's own error body, recovered at capture and carried BESIDE
+     * {@link content}, which stays exactly what the model observed. `tool_result_json_path` grades
+     * this when it is present; when it is absent the check reads {@link content} and behaves as it
+     * always has. Present implies it parses as JSON, so a check that finds it never has to ask.
+     *
+     * Absent for everything that is not an errored MCP call, and absent too for an MCP error whose
+     * body is prose rather than JSON — a server that puts its detail only in MCP `structuredContent`
+     * is not gradable at all, because the adapter discards that before any of our code runs.
+     */
+    errorPayload?: string;
 }
 /** What one attempt at running a cell through the shared single-shot runtime produced. */
 export interface CellRunOutcome {
@@ -68,6 +80,13 @@ export interface CellRunOutcome {
      * {@link tools} but un-deduped and carrying each result's error status + payload. Only the
      * in-process `gth-agent` runner can populate this (external adk-agent/ag-ui targets never do). */
     toolResults?: ToolResultRecord[];
+    /**
+     * BATCH-32 — what the agent ADVERTISED to the model for this cell: the tool-coverage denominator,
+     * to {@link tools}'s numerator. Only the in-process `gth-agent` runner can populate it (an
+     * external adk-agent/ag-ui target never advertises its inventory over the wire), and a test fake
+     * leaves it unset — which is why absence means "no denominator available", never "no tools".
+     */
+    advertisedTools?: AdvertisedToolInventory;
     /** A human-readable failure reason, set when `ok` is `false`. */
     error?: string;
 }

package/dist/types.js

@@ -1,12 +1,3 @@
-/**
- * @packageDocumentation
- * BATCH-1 — the shapes shared across matrix construction, the concurrency/retry runner, and the
- * structured output writer. Deliberately independent of any LLM/runner types (`GthConfig`,
- * `runSingleShot`, …): this package only knows about *cells* and *outcomes*. The production
- * adapter that turns a cell into an actual `runSingleShot` call lives in
- * `packages/app/src/commands/batchCommand.ts`, which is what keeps {@link RunCellFn} injectable —
- * unit tests here fake it, the CLI wires the real one.
- */
 /**
  * BATCH-24 — the default **cell** concurrency cap for `gth batch` / `gth eval` when
  * `-j/--concurrency` is not supplied. Serial by default, for every provider (deliberately NOT

package/dist/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAkHH;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,wBAAwB,GAAG,CAAC,CAAC;AAE1C;;;;GAIG;AACH,MAAM,CAAC,MAAM,4BAA4B,GAAG,CAAC,CAAC"}
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AA6IA;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,wBAAwB,GAAG,CAAC,CAAC;AAE1C;;;;GAIG;AACH,MAAM,CAAC,MAAM,4BAA4B,GAAG,CAAC,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gaunt-sloth/batch",
-  "version": "2.0.0-beta.10",
+  "version": "2.0.0-beta.12",
   "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",
@@ -37,8 +37,8 @@
     "@langchain/core": "^1.2.11",
     "yaml": "^2.9.1",
     "zod": "^4.6.3",
-    "@gaunt-sloth/agent": "2.0.0-beta.10",
-    "@gaunt-sloth/core": "2.0.0-beta.10"
+    "@gaunt-sloth/core": "2.0.0-beta.12",
+    "@gaunt-sloth/agent": "2.0.0-beta.12"
   },
   "files": [
     "./dist/*"