@gaunt-sloth/core 2.0.0-alpha.19 → 2.0.0-alpha.20

package/dist/providers/ollama.js

@@ -3,58 +3,76 @@ import { env } from '#src/utils/systemUtils.js';
 import { writeConfigFileWithMessages } from '#src/utils/fileUtils.js';
 import { buildInitConfigContent, getCuratedFallbackModel } from '#src/providers/modelDiscovery.js';
 /**
- * Default Ollama daemon host, matching the Ollama CLI/library default. The
- * OpenAI-compatible surface lives under `/v1` on this host. Kept in sync with
- * `DEFAULT_OLLAMA_HOST` in `modelDiscovery.ts`.
+ * Default Ollama daemon host, matching the Ollama CLI/library default. GS2-59 — `ChatOllama` talks
+ * to the daemon's NATIVE endpoint (`/api/chat`), NOT the OpenAI-compatible `/v1` shim this provider
+ * used before. The same root host is reused by `modelDiscovery.ts`'s model picker, which still
+ * queries the `/v1/models` surface (discovery only); both derive from this root — they stay in sync
+ * at the host level even though the two use different endpoints under it.
  */
 const DEFAULT_OLLAMA_HOST = 'http://127.0.0.1:11434';
 /**
- * Ollama serves an unauthenticated local daemon, but `ChatOpenAI` requires a
- * non-empty `apiKey` string. Send a harmless placeholder so the client builds;
- * the local daemon ignores it.
+ * GS2-59 — default context window (`num_ctx`) for Ollama models. Ollama's OWN default is 4096, but
+ * gaunt-sloth's agentic prompt (system + full lean toolset + a tool result) already lands ~4000
+ * tokens; at 4096 a thinking model (e.g. gemma4:31b) spends its entire remaining budget on the
+ * reasoning field and emits EMPTY `content` on the turn after a tool executes — the GS2-59
+ * blank-answer regression. The previous `ChatOpenAI`→`/v1` path could not fix this because the
+ * OpenAI-compat shim IGNORES `num_ctx`; the native `/api/chat` path honors it.
+ *
+ * 16384 is chosen as the largest window that is BOTH safely above the ~4000-token starvation point
+ * (4× headroom for reasoning + a few tool results) AND fits constrained consumer VRAM: Ollama
+ * preallocates the KV cache at `num_ctx`, so on a box where a large model already spills partly to
+ * CPU (e.g. a 19GB model on ~18GB of GPU), a 32768 cache tips the GPU allocation into an
+ * out-of-memory error. 16384 was verified live to run the agentic tool→synthesis turn on such a
+ * box; 32768 OOM'd it. Overridable per config via `llm.numCtx` — raise it if you have the VRAM and
+ * run long sessions, lower it on very tight hardware. NOTE: a per-request `num_ctx` overrides the
+ * daemon's `OLLAMA_CONTEXT_LENGTH`, so a user who tuned their server window higher should set
+ * `llm.numCtx` to match rather than rely on the server default.
  */
-const OLLAMA_PLACEHOLDER_API_KEY = 'ollama';
+const DEFAULT_OLLAMA_NUM_CTX = 16384;
 /**
- * Resolve the OpenAI-compatible base URL for the local Ollama daemon.
- *
- * Honors the `OLLAMA_HOST` env override (the same variable the Ollama CLI uses).
- * `OLLAMA_HOST` is typically a full URL (`http://127.0.0.1:11434`) or a bare
- * `host:port`; either form is normalized to a `http(s)://host[:port]/v1` base.
+ * Resolve the base URL for the local Ollama daemon — the NATIVE root (no `/v1` suffix; `ChatOllama`
+ * appends its own `/api/chat` path). Honors the `OLLAMA_HOST` env override (the same variable the
+ * Ollama CLI uses): a full URL (`http://127.0.0.1:11434`) or a bare `host:port` is normalized to
+ * `http(s)://host[:port]` with any trailing slash stripped.
  */
 function resolveBaseUrl() {
     const host = env.OLLAMA_HOST;
-    let base;
-    if (!host) {
-        base = DEFAULT_OLLAMA_HOST;
-    }
-    else if (/^https?:\/\//.test(host)) {
-        base = host.replace(/\/+$/, '');
-    }
-    else {
-        base = `http://${host}`.replace(/\/+$/, '');
-    }
-    return `${base}/v1`;
+    if (!host)
+        return DEFAULT_OLLAMA_HOST;
+    const base = /^https?:\/\//.test(host) ? host : `http://${host}`;
+    return base.replace(/\/+$/, '');
 }
-// Function to process JSON config and create an Ollama (OpenAI-compatible) LLM instance
+/**
+ * GS2-59 — build a native `ChatOllama` client (`@langchain/ollama`, talking to Ollama's
+ * `/api/chat`). This replaces the previous `ChatOpenAI`→`/v1` client, which had two fatal
+ * properties for local thinking models: it DROPPED the model's separate reasoning field (the
+ * completions converter discards it — `__includeRawResponse` is openrouter-only), and it IGNORED
+ * `num_ctx` so a large agentic prompt starved the answer to empty content. `ChatOllama` fixes both:
+ * it honors `numCtx`, and it surfaces the model's thinking in `additional_kwargs.reasoning_content`
+ * — the exact field the reasoning pipeline (`pickReasoningDelta`) already reads — so the `/reasoning`
+ * panel populates with no downstream change.
+ *
+ * Ollama is a local, unauthenticated daemon, so the old OpenAI-client knobs are gone (2.0 migration
+ * note): point elsewhere with `OLLAMA_HOST`; the previous `configuration.baseURL` / placeholder
+ * `apiKey` fields no longer apply. If a reverse proxy in front of Ollama needs auth, set `headers`
+ * in the config (passed straight through to `ChatOllama`).
+ */
 // noinspection JSUnusedGlobalSymbols
 export async function processJsonConfig(llmConfig) {
-    const { ChatOpenAI } = await import('@langchain/openai');
-    // Ollama is local and unauthenticated; ChatOpenAI still needs a non-empty key.
-    const apiKey = llmConfig.apiKey || OLLAMA_PLACEHOLDER_API_KEY;
+    const { ChatOllama } = await import('@langchain/ollama');
     const configFields = {
         ...llmConfig,
-        apiKey,
         model: llmConfig.model || getCuratedFallbackModel('ollama'),
-        configuration: {
-            baseURL: resolveBaseUrl(),
-            ...(llmConfig.configuration || {}),
-        },
+        // Precedence: an explicit config `baseUrl` (the native ChatOllama field) wins, else the
+        // `OLLAMA_HOST` env, else the local default. Config is the more specific/intentional signal.
+        baseUrl: llmConfig.baseUrl ?? resolveBaseUrl(),
+        numCtx: llmConfig.numCtx ?? DEFAULT_OLLAMA_NUM_CTX,
     };
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    delete configFields.type;
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    delete configFields.apiKeyEnvironmentVariable;
-    return new ChatOpenAI(configFields);
+    // Strip OpenAI-client / gaunt-sloth-internal keys ChatOllama neither needs nor understands.
+    for (const key of ['type', 'apiKeyEnvironmentVariable', 'apiKey', 'configuration']) {
+        delete configFields[key];
+    }
+    return new ChatOllama(configFields);
 }
 export function init(configFileName, force = false, model) {
     // Determine which content to use based on file extension

package/dist/providers/ollama.js.map

@@ -1 +1 @@
-{"version":3,"file":"ollama.js","sourceRoot":"","sources":["../../src/providers/ollama.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAQhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AAEnG;;;;GAIG;AACH,MAAM,mBAAmB,GAAG,wBAAwB,CAAC;AAErD;;;;GAIG;AACH,MAAM,0BAA0B,GAAG,QAAQ,CAAC;AAE5C;;;;;;GAMG;AACH,SAAS,cAAc;IACrB,MAAM,IAAI,GAAG,GAAG,CAAC,WAAW,CAAC;IAC7B,IAAI,IAAY,CAAC;IACjB,IAAI,CAAC,IAAI,EAAE,CAAC;QACV,IAAI,GAAG,mBAAmB,CAAC;IAC7B,CAAC;SAAM,IAAI,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QACrC,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IAClC,CAAC;SAAM,CAAC;QACN,IAAI,GAAG,UAAU,IAAI,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IAC9C,CAAC;IACD,OAAO,GAAG,IAAI,KAAK,CAAC;AACtB,CAAC;AAED,wFAAwF;AACxF,qCAAqC;AACrC,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAmE;IAEnE,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACzD,+EAA+E;IAC/E,MAAM,MAAM,GAAG,SAAS,CAAC,MAAM,IAAI,0BAA0B,CAAC;IAC9D,MAAM,YAAY,GAAG;QACnB,GAAG,SAAS;QACZ,MAAM;QACN,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,QAAQ,CAAC;QAC3D,aAAa,EAAE;YACb,OAAO,EAAE,cAAc,EAAE;YACzB,GAAG,CAAC,SAAS,CAAC,aAAa,IAAI,EAAE,CAAC;SACnC;KACF,CAAC;IACF,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,IAAI,CAAC;IAClC,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,yBAAyB,CAAC;IAEvD,OAAO,IAAI,UAAU,CAAC,YAAY,CAAC,CAAC;AACtC,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAC5F,cAAc,CACZ,yBAAyB,cAAc,2BAA2B;QAChE,2EAA2E;QAC3E,+CAA+C,CAClD,CAAC;AACJ,CAAC"}
+{"version":3,"file":"ollama.js","sourceRoot":"","sources":["../../src/providers/ollama.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAOhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AAEnG;;;;;;GAMG;AACH,MAAM,mBAAmB,GAAG,wBAAwB,CAAC;AAErD;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,sBAAsB,GAAG,KAAK,CAAC;AAErC;;;;;GAKG;AACH,SAAS,cAAc;IACrB,MAAM,IAAI,GAAG,GAAG,CAAC,WAAW,CAAC;IAC7B,IAAI,CAAC,IAAI;QAAE,OAAO,mBAAmB,CAAC;IACtC,MAAM,IAAI,GAAG,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,IAAI,EAAE,CAAC;IACjE,OAAO,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;AAClC,CAAC;AAED;;;;;;;;;;;;;;GAcG;AACH,qCAAqC;AACrC,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAgD;IAEhD,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACzD,MAAM,YAAY,GAA4B;QAC5C,GAAG,SAAS;QACZ,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,QAAQ,CAAC;QAC3D,wFAAwF;QACxF,6FAA6F;QAC7F,OAAO,EAAE,SAAS,CAAC,OAAO,IAAI,cAAc,EAAE;QAC9C,MAAM,EAAE,SAAS,CAAC,MAAM,IAAI,sBAAsB;KACnD,CAAC;IACF,4FAA4F;IAC5F,KAAK,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,2BAA2B,EAAE,QAAQ,EAAE,eAAe,CAAC,EAAE,CAAC;QACnF,OAAO,YAAY,CAAC,GAAG,CAAC,CAAC;IAC3B,CAAC;IACD,OAAO,IAAI,UAAU,CAAC,YAA+B,CAAC,CAAC;AACzD,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAC5F,cAAc,CACZ,yBAAyB,cAAc,2BAA2B;QAChE,2EAA2E;QAC3E,+CAA+C,CAClD,CAAC;AACJ,CAAC"}

package/dist/providers/openrouter.js

@@ -15,6 +15,11 @@ export async function processJsonConfig(llmConfig) {
         ...llmConfig,
         apiKey: openRouterApiKey,
         model: llmConfig.model || getCuratedFallbackModel('openrouter'),
+        // TUI-C22 — OpenRouter returns a thinking model's reasoning in a top-level `reasoning` field
+        // that the ChatOpenAI completions converter drops. `__includeRawResponse` stashes the raw
+        // provider response under `additional_kwargs.__raw_response`, which GthAbstractAgent reads
+        // (`choices[0].delta.reasoning`) to populate the /reasoning panel. Kept opt-out via config.
+        __includeRawResponse: llmConfig.__includeRawResponse ?? true,
         configuration: {
             baseURL: 'https://openrouter.ai/api/v1',
             ...(llmConfig.configuration || {}),

package/dist/providers/openrouter.js.map

@@ -1 +1 @@
-{"version":3,"file":"openrouter.js","sourceRoot":"","sources":["../../src/providers/openrouter.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAQhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AAEnG,qEAAqE;AACrE,qCAAqC;AACrC,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAmE;IAEnE,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACzD,wEAAwE;IACxE,MAAM,gBAAgB,GAAG,SAAS,CAAC,SAAS,CAAC,CAAC;IAC9C,IAAI,CAAC,gBAAgB,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CACb,iGAAiG,CAClG,CAAC;IACJ,CAAC;IACD,MAAM,YAAY,GAAG;QACnB,GAAG,SAAS;QACZ,MAAM,EAAE,gBAAgB;QACxB,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,YAAY,CAAC;QAC/D,aAAa,EAAE;YACb,OAAO,EAAE,8BAA8B;YACvC,GAAG,CAAC,SAAS,CAAC,aAAa,IAAI,EAAE,CAAC;YAClC,cAAc,EAAE;gBACd,cAAc,EAAE,yBAAyB;gBACzC,SAAS,EAAE,aAAa;aACzB;SACF;KACF,CAAC;IACF,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,IAAI,CAAC;IAClC,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,yBAAyB,CAAC;IACvD,OAAO,IAAI,UAAU,CAAC,YAAY,CAAC,CAAC;AACtC,CAAC;AAED,SAAS,SAAS,CAAC,SAAmE;IACpF,8DAA8D;IAC9D,MAAM,IAAI,GAAG,SAA0C,CAAC;IACxD,IAAI,IAAI,CAAC,yBAAyB,IAAI,GAAG,CAAC,IAAI,CAAC,yBAAyB,CAAC,EAAE,CAAC;QAC1E,OAAO,GAAG,CAAC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAC7C,CAAC;SAAM,CAAC;QACN,OAAO,SAAS,CAAC,MAAM,IAAI,GAAG,CAAC,mBAAmB,IAAI,GAAG,CAAC,kBAAkB,CAAC;IAC/E,CAAC;AACH,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,YAAY,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAChG,cAAc,CACZ,yBAAyB,cAAc,uBAAuB;QAC5D,qDAAqD,CACxD,CAAC;AACJ,CAAC"}
+{"version":3,"file":"openrouter.js","sourceRoot":"","sources":["../../src/providers/openrouter.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAQhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AAEnG,qEAAqE;AACrE,qCAAqC;AACrC,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAmE;IAEnE,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACzD,wEAAwE;IACxE,MAAM,gBAAgB,GAAG,SAAS,CAAC,SAAS,CAAC,CAAC;IAC9C,IAAI,CAAC,gBAAgB,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CACb,iGAAiG,CAClG,CAAC;IACJ,CAAC;IACD,MAAM,YAAY,GAAG;QACnB,GAAG,SAAS;QACZ,MAAM,EAAE,gBAAgB;QACxB,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,YAAY,CAAC;QAC/D,6FAA6F;QAC7F,0FAA0F;QAC1F,2FAA2F;QAC3F,4FAA4F;QAC5F,oBAAoB,EAAE,SAAS,CAAC,oBAAoB,IAAI,IAAI;QAC5D,aAAa,EAAE;YACb,OAAO,EAAE,8BAA8B;YACvC,GAAG,CAAC,SAAS,CAAC,aAAa,IAAI,EAAE,CAAC;YAClC,cAAc,EAAE;gBACd,cAAc,EAAE,yBAAyB;gBACzC,SAAS,EAAE,aAAa;aACzB;SACF;KACF,CAAC;IACF,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,IAAI,CAAC;IAClC,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,yBAAyB,CAAC;IACvD,OAAO,IAAI,UAAU,CAAC,YAAY,CAAC,CAAC;AACtC,CAAC;AAED,SAAS,SAAS,CAAC,SAAmE;IACpF,8DAA8D;IAC9D,MAAM,IAAI,GAAG,SAA0C,CAAC;IACxD,IAAI,IAAI,CAAC,yBAAyB,IAAI,GAAG,CAAC,IAAI,CAAC,yBAAyB,CAAC,EAAE,CAAC;QAC1E,OAAO,GAAG,CAAC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAC7C,CAAC;SAAM,CAAC;QACN,OAAO,SAAS,CAAC,MAAM,IAAI,GAAG,CAAC,mBAAmB,IAAI,GAAG,CAAC,kBAAkB,CAAC;IAC/E,CAAC;AACH,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,YAAY,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAChG,cAAc,CACZ,yBAAyB,cAAc,uBAAuB;QAC5D,qDAAqD,CACxD,CAAC;AACJ,CAAC"}

package/dist/providers/vertexai.js

@@ -13,6 +13,7 @@
 import { displayWarning } from '#src/utils/consoleUtils.js';
 import { writeConfigFileWithMessages } from '#src/utils/fileUtils.js';
 import { buildInitConfigContent, getCuratedFallbackModel } from '#src/providers/modelDiscovery.js';
+import { applyGeminiToolSchemaSanitizer } from '#src/providers/geminiSchemaSanitizer.js';
 export function init(configFileName, force = false, model) {
     // Determine which content to use based on file extension
     if (!configFileName.endsWith('.json')) {
@@ -31,6 +32,8 @@ export async function processJsonConfig(llmConfig) {
     };
     delete configFields.type;
     delete configFields.apiKeyEnvironmentVariable;
-    return new ChatGoogle(configFields);
+    // GS2-58: normalise every tool's JSON-Schema at the ChatGoogle boundary so Gemini's OpenAPI-3.0
+    // subset accepts built-in, custom, and MCP tools alike (see geminiSchemaSanitizer).
+    return applyGeminiToolSchemaSanitizer(new ChatGoogle(configFields));
 }
 //# sourceMappingURL=vertexai.js.map

package/dist/providers/vertexai.js.map

@@ -1 +1 @@
-{"version":3,"file":"vertexai.js","sourceRoot":"","sources":["../../src/providers/vertexai.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AACH,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAI5D,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AAEnG,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,UAAU,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAC9F,cAAc,CACZ,+GAA+G,CAChH,CAAC;AACJ,CAAC;AAED,mEAAmE;AACnE,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAmF;IAEnF,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,wBAAwB,CAAC,CAAC;IAC9D,MAAM,YAAY,GAAG;QACnB,GAAG,SAAS;QACZ,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,UAAU,CAAC;QAC7D,QAAQ,EAAE,IAAI;KACf,CAAC;IACF,OAAO,YAAY,CAAC,IAAI,CAAC;IACzB,OAAO,YAAY,CAAC,yBAAyB,CAAC;IAC9C,OAAO,IAAI,UAAU,CAAC,YAAY,CAAC,CAAC;AACtC,CAAC"}
+{"version":3,"file":"vertexai.js","sourceRoot":"","sources":["../../src/providers/vertexai.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AACH,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAI5D,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AACnG,OAAO,EAAE,8BAA8B,EAAE,MAAM,yCAAyC,CAAC;AAEzF,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,UAAU,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAC9F,cAAc,CACZ,+GAA+G,CAChH,CAAC;AACJ,CAAC;AAED,mEAAmE;AACnE,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAmF;IAEnF,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,wBAAwB,CAAC,CAAC;IAC9D,MAAM,YAAY,GAAG;QACnB,GAAG,SAAS;QACZ,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,UAAU,CAAC;QAC7D,QAAQ,EAAE,IAAI;KACf,CAAC;IACF,OAAO,YAAY,CAAC,IAAI,CAAC;IACzB,OAAO,YAAY,CAAC,yBAAyB,CAAC;IAC9C,gGAAgG;IAChG,oFAAoF;IACpF,OAAO,8BAA8B,CAAC,IAAI,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC;AACtE,CAAC"}

package/dist/runtime/conversation.d.ts

@@ -0,0 +1,59 @@
+import type { GthConfig } from '#src/config.js';
+import type { AgentResolvers, GthAgentFactory, GthCommand } from '#src/core/types.js';
+import type { GthRunStats } from '#src/core/types.js';
+/**
+ * One turn's result inside a {@link runConversation} run: the per-turn `ok`/`answer` plus that
+ * turn's run stats (GS2-16 {@link GthRunStats} — token usage + invoked tools), captured PER TURN (a
+ * per-invoke delta, not the cumulative conversation total). `ok` is `false` when that turn's agent
+ * invocation failed (`error` set, `answer` empty). Extends `GthRunStats` rather than restating its
+ * fields — same shape {@link ../runtime/singleShot.js SingleShotResult} uses.
+ */
+export interface ConversationTurnResult extends GthRunStats {
+    /** `true` when this turn completed without error, `false` when it failed. */
+    ok: boolean;
+    /** This turn's full answer text (`runner.processMessages()`'s return value). Empty on failure. */
+    answer: string;
+    /** Set when `ok` is `false`: why this turn failed. */
+    error?: string;
+}
+/**
+ * Run a scripted MULTI-TURN conversation and return one {@link ConversationTurnResult} per turn.
+ *
+ * This is the **conversational** counterpart to {@link ../runtime/singleShot.js runSingleShot}
+ * (which is stateless — a fresh agent per call). It builds the agent + resolves tools ONCE, then
+ * runs each turn against the ACCUMULATED message history so cross-turn "memory" / identity behaviour
+ * is real, and cleans up ONCE at the end (reusing runSingleShot's cleanup discipline — the resolvers
+ * are the caller's to tear down, exactly as with runSingleShot).
+ *
+ * **History mechanism = stateless replay of the growing message array.** Messages accumulate as
+ * `[user1, ai1, user2, ai2, …]`: per turn a `HumanMessage(user)` is appended, the agent runs on the
+ * WHOLE array, and its answer is appended as an `AIMessage` so the next turn sees it. The system
+ * prompt is NOT seeded here — the agent composes it via `createAgent({ systemPrompt })` (BATCH-13;
+ * see `runSingleShot`), so the replayed array carries only human/assistant turns. Before each
+ * turn the runner's thread is rotated ({@link GthAgentRunner.resetThread}) so the checkpointer starts
+ * empty and the replayed array is the sole history (no `add_messages` double-append). This mirrors
+ * the AG-UI server's "client is the source of truth for history — it sends the full message list
+ * every turn" model and reuses the existing `processMessages` + `resetThread` machinery with no new
+ * agent surface. **Known limitation (unverified pending a live pass):** replay carries prior
+ * *answers* (as `AIMessage` text) but NOT prior tool-call / tool-result messages — a checkpointer-
+ * thread approach (send only the new message, let `add_messages` accumulate) would preserve those.
+ *
+ * **Per-turn tool capture (GS2-16):** `processMessages` resets the analytics tally at its top, so
+ * `getRunStats()` read right after each turn returns THAT turn's tool/token delta (not cumulative).
+ *
+ * A turn that fails is recorded (`ok:false`, `error`) and the conversation STOPS (later turns depend
+ * on the broken context), so the returned array may be shorter than `userMessages` — the caller
+ * (`gth eval`'s runner) fails the un-run turns.
+ *
+ * @param source - The source label (used for output/session-file naming), e.g. `EVAL-<cellId>`.
+ * @param _preamble - Deprecated/ignored (BATCH-13): the agent composes the system prompt itself (via
+ *   `createAgent({ systemPrompt })`); seeding it here too produced a second system message that
+ *   `@langchain/anthropic` rejects. Retained positionally so existing callers need no change.
+ * @param userMessages - The ordered user turns to send (one conversation).
+ * @param config - The resolved config.
+ * @param resolvers - Optional agent resolvers (tools/middleware); the caller owns their cleanup.
+ * @param command - The originating command (defaults to `ask`); selects the agent mode prompt.
+ * @param agentFactory - Optional backend factory (B5); omitted = the runner's lean default.
+ * @returns One {@link ConversationTurnResult} per turn attempted, in turn order.
+ */
+export declare function runConversation(source: string, _preamble: string, userMessages: string[], config: GthConfig, resolvers?: AgentResolvers, command?: GthCommand, agentFactory?: GthAgentFactory): Promise<ConversationTurnResult[]>;

package/dist/runtime/conversation.js

@@ -0,0 +1,137 @@
+import { defaultStatusCallback, display, displayError, displaySuccess, flushSessionLog, initSessionLogging, stopSessionLogging, } from '#src/utils/consoleUtils.js';
+import { getCommandOutputFilePath } from '#src/utils/fileUtils.js';
+import { GthAgentRunner } from '#src/core/GthAgentRunner.js';
+import { MemorySaver } from '@langchain/langgraph';
+import { AIMessage, HumanMessage } from '@langchain/core/messages';
+import { ProgressIndicator } from '#src/utils/ProgressIndicator.js';
+import { recordSessionSafe } from '#src/history/recordSession.js';
+import { getProjectDir } from '#src/utils/systemUtils.js';
+/**
+ * Run a scripted MULTI-TURN conversation and return one {@link ConversationTurnResult} per turn.
+ *
+ * This is the **conversational** counterpart to {@link ../runtime/singleShot.js runSingleShot}
+ * (which is stateless — a fresh agent per call). It builds the agent + resolves tools ONCE, then
+ * runs each turn against the ACCUMULATED message history so cross-turn "memory" / identity behaviour
+ * is real, and cleans up ONCE at the end (reusing runSingleShot's cleanup discipline — the resolvers
+ * are the caller's to tear down, exactly as with runSingleShot).
+ *
+ * **History mechanism = stateless replay of the growing message array.** Messages accumulate as
+ * `[user1, ai1, user2, ai2, …]`: per turn a `HumanMessage(user)` is appended, the agent runs on the
+ * WHOLE array, and its answer is appended as an `AIMessage` so the next turn sees it. The system
+ * prompt is NOT seeded here — the agent composes it via `createAgent({ systemPrompt })` (BATCH-13;
+ * see `runSingleShot`), so the replayed array carries only human/assistant turns. Before each
+ * turn the runner's thread is rotated ({@link GthAgentRunner.resetThread}) so the checkpointer starts
+ * empty and the replayed array is the sole history (no `add_messages` double-append). This mirrors
+ * the AG-UI server's "client is the source of truth for history — it sends the full message list
+ * every turn" model and reuses the existing `processMessages` + `resetThread` machinery with no new
+ * agent surface. **Known limitation (unverified pending a live pass):** replay carries prior
+ * *answers* (as `AIMessage` text) but NOT prior tool-call / tool-result messages — a checkpointer-
+ * thread approach (send only the new message, let `add_messages` accumulate) would preserve those.
+ *
+ * **Per-turn tool capture (GS2-16):** `processMessages` resets the analytics tally at its top, so
+ * `getRunStats()` read right after each turn returns THAT turn's tool/token delta (not cumulative).
+ *
+ * A turn that fails is recorded (`ok:false`, `error`) and the conversation STOPS (later turns depend
+ * on the broken context), so the returned array may be shorter than `userMessages` — the caller
+ * (`gth eval`'s runner) fails the un-run turns.
+ *
+ * @param source - The source label (used for output/session-file naming), e.g. `EVAL-<cellId>`.
+ * @param _preamble - Deprecated/ignored (BATCH-13): the agent composes the system prompt itself (via
+ *   `createAgent({ systemPrompt })`); seeding it here too produced a second system message that
+ *   `@langchain/anthropic` rejects. Retained positionally so existing callers need no change.
+ * @param userMessages - The ordered user turns to send (one conversation).
+ * @param config - The resolved config.
+ * @param resolvers - Optional agent resolvers (tools/middleware); the caller owns their cleanup.
+ * @param command - The originating command (defaults to `ask`); selects the agent mode prompt.
+ * @param agentFactory - Optional backend factory (B5); omitted = the runner's lean default.
+ * @returns One {@link ConversationTurnResult} per turn attempted, in turn order.
+ */
+export async function runConversation(source, _preamble, userMessages, config, resolvers, command = 'ask', agentFactory) {
+    const progressIndicator = config.streamOutput ? undefined : new ProgressIndicator('Thinking.');
+    // Resolve output path and initialize session logging if enabled (same discipline as runSingleShot;
+    // a no-op when `writeOutputToFile` is off, as `gth eval` forces it — getCommandOutputFilePath null).
+    const filePath = getCommandOutputFilePath(config, source);
+    if (filePath) {
+        initSessionLogging(filePath, config.streamSessionInferenceLog);
+    }
+    // Build the agent + resolve tools ONCE for the whole conversation (the MCP connection / any OAuth /
+    // the toolset must persist across turns so cross-turn memory is real). Cleaned up once, in finally.
+    const runner = new GthAgentRunner(defaultStatusCallback, resolvers, agentFactory);
+    const results = [];
+    // The accumulated conversation: [user1, ai1, user2, ai2, …]. Each turn replays the whole array
+    // against a freshly-rotated thread (see the doc block). BATCH-13: NO leading SystemMessage — the
+    // agent composes the system prompt via `createAgent({ systemPrompt })` (same as runSingleShot);
+    // seeding a preamble SystemMessage here too made two system messages, which Anthropic rejects.
+    const messages = [];
+    try {
+        await runner.init(command, config, new MemorySaver());
+        for (const userMessage of userMessages) {
+            // Rotate to a fresh (empty) checkpointer thread so this turn's replay of the full `messages`
+            // array is the sole history the agent sees — no double-append from a prior turn's checkpoint.
+            runner.resetThread();
+            messages.push(new HumanMessage(userMessage));
+            const startedAt = Date.now();
+            let answer = '';
+            let ok = true;
+            let error;
+            try {
+                answer = await runner.processMessages(messages);
+                // Append this turn's answer so the NEXT turn's replay includes it (cross-turn memory).
+                messages.push(new AIMessage(answer));
+            }
+            catch (err) {
+                ok = false;
+                error = err instanceof Error ? err.message : String(err);
+                displayError(`Failed to get answer: ${error}`);
+            }
+            // GS2-16: read this turn's token/tool delta from the live agent (before cleanup). Fail-soft —
+            // analytics must never affect the run. `processMessages` reset the tally at its top, so this is
+            // THIS turn's usage, not the conversation's cumulative total.
+            let runStats = { tools: [] };
+            try {
+                const s = runner.getRunStats?.();
+                if (s)
+                    runStats = s;
+            }
+            catch {
+                /* fail-soft */
+            }
+            // GS2-7 (B20): opt-in, fail-soft per-turn session history. A no-op unless `history.enabled`.
+            recordSessionSafe(config, {
+                command,
+                project: getProjectDir(),
+                model: config.modelDisplayName,
+                prompt: userMessage,
+                response: answer,
+                tokensInput: runStats.tokensInput,
+                tokensOutput: runStats.tokensOutput,
+                tools: runStats.tools.length > 0 ? runStats.tools : undefined,
+                durationMs: Date.now() - startedAt,
+            });
+            results.push({ ok, answer, error, ...runStats });
+            // A failed turn breaks the conversation's context — stop rather than run later turns on it.
+            if (!ok)
+                break;
+        }
+    }
+    finally {
+        await runner.cleanup();
+    }
+    progressIndicator?.stop();
+    if (config.writeOutputToFile === false) {
+        display('\n'); // something going on in some terminals, they swallow last line of output
+    }
+    if (filePath) {
+        try {
+            flushSessionLog();
+            stopSessionLogging();
+            displaySuccess(`\n\nThis report can be found in ${filePath}`);
+        }
+        catch (err) {
+            displayError(`Failed to write answer to file: ${filePath}`);
+            displayError(err instanceof Error ? err.message : String(err));
+        }
+    }
+    return results;
+}
+//# sourceMappingURL=conversation.js.map

package/dist/runtime/conversation.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"conversation.js","sourceRoot":"","sources":["../../src/runtime/conversation.ts"],"names":[],"mappings":"AACA,OAAO,EACL,qBAAqB,EACrB,OAAO,EACP,YAAY,EACZ,cAAc,EACd,eAAe,EACf,kBAAkB,EAClB,kBAAkB,GACnB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AACnE,OAAO,EAAE,cAAc,EAAE,MAAM,6BAA6B,CAAC;AAC7D,OAAO,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AACnD,OAAO,EAAE,SAAS,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAEnE,OAAO,EAAE,iBAAiB,EAAE,MAAM,iCAAiC,CAAC;AAEpE,OAAO,EAAE,iBAAiB,EAAE,MAAM,+BAA+B,CAAC;AAElE,OAAO,EAAE,aAAa,EAAE,MAAM,2BAA2B,CAAC;AAkB1D;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAuCG;AACH,MAAM,CAAC,KAAK,UAAU,eAAe,CACnC,MAAc,EACd,SAAiB,EACjB,YAAsB,EACtB,MAAiB,EACjB,SAA0B,EAC1B,OAAO,GAAe,KAAK,EAC3B,YAA8B;IAE9B,MAAM,iBAAiB,GAAG,MAAM,CAAC,YAAY,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,iBAAiB,CAAC,WAAW,CAAC,CAAC;IAE/F,mGAAmG;IACnG,qGAAqG;IACrG,MAAM,QAAQ,GAAG,wBAAwB,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC1D,IAAI,QAAQ,EAAE,CAAC;QACb,kBAAkB,CAAC,QAAQ,EAAE,MAAM,CAAC,yBAAyB,CAAC,CAAC;IACjE,CAAC;IAED,oGAAoG;IACpG,oGAAoG;IACpG,MAAM,MAAM,GAAG,IAAI,cAAc,CAAC,qBAAqB,EAAE,SAAS,EAAE,YAAY,CAAC,CAAC;IAClF,MAAM,OAAO,GAA6B,EAAE,CAAC;IAC7C,+FAA+F;IAC/F,iGAAiG;IACjG,gGAAgG;IAChG,+FAA+F;IAC/F,MAAM,QAAQ,GAAc,EAAE,CAAC;IAE/B,IAAI,CAAC;QACH,MAAM,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,IAAI,WAAW,EAAE,CAAC,CAAC;QAEtD,KAAK,MAAM,WAAW,IAAI,YAAY,EAAE,CAAC;YACvC,6FAA6F;YAC7F,8FAA8F;YAC9F,MAAM,CAAC,WAAW,EAAE,CAAC;YACrB,QAAQ,CAAC,IAAI,CAAC,IAAI,YAAY,CAAC,WAAW,CAAC,CAAC,CAAC;YAE7C,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,IAAI,MAAM,GAAG,EAAE,CAAC;YAChB,IAAI,EAAE,GAAG,IAAI,CAAC;YACd,IAAI,KAAyB,CAAC;YAC9B,IAAI,CAAC;gBACH,MAAM,GAAG,MAAM,MAAM,CAAC,eAAe,CAAC,QAAQ,CAAC,CAAC;gBAChD,uFAAuF;gBACvF,QAAQ,CAAC,IAAI,CAAC,IAAI,SAAS,CAAC,MAAM,CAAC,CAAC,CAAC;YACvC,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,EAAE,GAAG,KAAK,CAAC;gBACX,KAAK,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;gBACzD,YAAY,CAAC,yBAAyB,KAAK,EAAE,CAAC,CAAC;YACjD,CAAC;YAED,8FAA8F;YAC9F,gGAAgG;YAChG,8DAA8D;YAC9D,IAAI,QAAQ,GAAgB,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC;YAC1C,IAAI,CAAC;gBACH,MAAM,CAAC,GAAG,MAAM,CAAC,WAAW,EAAE,EAAE,CAAC;gBACjC,IAAI,CAAC;oBAAE,QAAQ,GAAG,CAAC,CAAC;YACtB,CAAC;YAAC,MAAM,CAAC;gBACP,eAAe;YACjB,CAAC;YAED,6FAA6F;YAC7F,iBAAiB,CAAC,MAAM,EAAE;gBACxB,OAAO;gBACP,OAAO,EAAE,aAAa,EAAE;gBACxB,KAAK,EAAE,MAAM,CAAC,gBAAgB;gBAC9B,MAAM,EAAE,WAAW;gBACnB,QAAQ,EAAE,MAAM;gBAChB,WAAW,EAAE,QAAQ,CAAC,WAAW;gBACjC,YAAY,EAAE,QAAQ,CAAC,YAAY;gBACnC,KAAK,EAAE,QAAQ,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS;gBAC7D,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;aACnC,CAAC,CAAC;YAEH,OAAO,CAAC,IAAI,CAAC,EAAE,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,GAAG,QAAQ,EAAE,CAAC,CAAC;YAEjD,4FAA4F;YAC5F,IAAI,CAAC,EAAE;gBAAE,MAAM;QACjB,CAAC;IACH,CAAC;YAAS,CAAC;QACT,MAAM,MAAM,CAAC,OAAO,EAAE,CAAC;IACzB,CAAC;IAED,iBAAiB,EAAE,IAAI,EAAE,CAAC;IAE1B,IAAI,MAAM,CAAC,iBAAiB,KAAK,KAAK,EAAE,CAAC;QACvC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,yEAAyE;IAC1F,CAAC;IACD,IAAI,QAAQ,EAAE,CAAC;QACb,IAAI,CAAC;YACH,eAAe,EAAE,CAAC;YAClB,kBAAkB,EAAE,CAAC;YACrB,cAAc,CAAC,mCAAmC,QAAQ,EAAE,CAAC,CAAC;QAChE,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,YAAY,CAAC,mCAAmC,QAAQ,EAAE,CAAC,CAAC;YAC5D,YAAY,CAAC,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;QACjE,CAAC;IACH,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC"}

package/dist/runtime/singleShot.d.ts

@@ -21,7 +21,8 @@ export interface SingleShotResult extends GthRunStats {
  * is forwarded to the agent so it can pick the right mode prompt (e.g. exec-mode for `exec`).
  *
  * @param source - The source of the question (used for file naming)
- * @param preamble - The preamble to send to the LLM
+ * @param _preamble - Deprecated/ignored (BATCH-13): the agent composes the system prompt itself;
+ *   see the body comment. Retained positionally so existing callers need no change.
  * @param content - The content of the question
  * @param config - The resolved config
  * @param resolvers - Optional agent resolvers (tools/middleware)
@@ -34,4 +35,4 @@ export interface SingleShotResult extends GthRunStats {
  *   `tokensOutput`/`tools` carry the SUT's answer text and run stats for callers that need them
  *   (e.g. `gth batch`/`gth eval`).
  */
-export declare function runSingleShot(source: string, preamble: string, content: string, config: GthConfig, resolvers?: AgentResolvers, command?: GthCommand, agentFactory?: GthAgentFactory): Promise<SingleShotResult>;
+export declare function runSingleShot(source: string, _preamble: string, content: string, config: GthConfig, resolvers?: AgentResolvers, command?: GthCommand, agentFactory?: GthAgentFactory): Promise<SingleShotResult>;

package/dist/runtime/singleShot.js

@@ -2,7 +2,7 @@ import { defaultStatusCallback, display, displayError, displaySuccess, flushSess
 import { getCommandOutputFilePath } from '#src/utils/fileUtils.js';
 import { GthAgentRunner } from '#src/core/GthAgentRunner.js';
 import { MemorySaver } from '@langchain/langgraph';
-import { HumanMessage, SystemMessage } from '@langchain/core/messages';
+import { HumanMessage } from '@langchain/core/messages';
 import { ProgressIndicator } from '#src/utils/ProgressIndicator.js';
 import { recordSessionSafe } from '#src/history/recordSession.js';
 import { getProjectDir } from '#src/utils/systemUtils.js';
@@ -14,7 +14,8 @@ import { getProjectDir } from '#src/utils/systemUtils.js';
  * is forwarded to the agent so it can pick the right mode prompt (e.g. exec-mode for `exec`).
  *
  * @param source - The source of the question (used for file naming)
- * @param preamble - The preamble to send to the LLM
+ * @param _preamble - Deprecated/ignored (BATCH-13): the agent composes the system prompt itself;
+ *   see the body comment. Retained positionally so existing callers need no change.
  * @param content - The content of the question
  * @param config - The resolved config
  * @param resolvers - Optional agent resolvers (tools/middleware)
@@ -27,9 +28,20 @@ import { getProjectDir } from '#src/utils/systemUtils.js';
  *   `tokensOutput`/`tools` carry the SUT's answer text and run stats for callers that need them
  *   (e.g. `gth batch`/`gth eval`).
  */
-export async function runSingleShot(source, preamble, content, config, resolvers, command = 'ask', agentFactory) {
+export async function runSingleShot(source, 
+// BATCH-13: `_preamble` is retained for signature stability but is NO LONGER injected as a
+// SystemMessage. The agent backends (lean `GthLangChainAgent` since GS2-21, and `GthDeepAgent`)
+// each COMPOSE the full system prompt themselves from the same config — backstory + guidelines +
+// per-command mode prompt + system prompt, PLUS the model-identity (GS2-34) and MCP-instructions
+// (EXT-32) notes — and hand it to `createAgent` as `systemPrompt`. Also passing this preamble as a
+// leading SystemMessage produced TWO system messages, which `@langchain/anthropic` rejects
+// ("System messages are only permitted as the first passed message"), breaking EVERY single-shot
+// run (ask/exec/batch/eval) on Anthropic on BOTH backends (Google/OpenAI silently merged them).
+// The agent's composed prompt is a superset of this preamble, so dropping it is content-preserving.
+_preamble, content, config, resolvers, command = 'ask', agentFactory) {
     const progressIndicator = config.streamOutput ? undefined : new ProgressIndicator('Thinking.');
-    const messages = [new SystemMessage(preamble), new HumanMessage(content)];
+    // Only the human turn: the agent supplies the system prompt via `createAgent({ systemPrompt })`.
+    const messages = [new HumanMessage(content)];
     // Resolve output path and initialize session logging if enabled
     const filePath = getCommandOutputFilePath(config, source);
     if (filePath) {

package/dist/runtime/singleShot.js.map

@@ -1 +1 @@
-{"version":3,"file":"singleShot.js","sourceRoot":"","sources":["../../src/runtime/singleShot.ts"],"names":[],"mappings":"AACA,OAAO,EACL,qBAAqB,EACrB,OAAO,EACP,YAAY,EACZ,cAAc,EACd,eAAe,EACf,kBAAkB,EAClB,kBAAkB,GACnB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AACnE,OAAO,EAAE,cAAc,EAAE,MAAM,6BAA6B,CAAC;AAC7D,OAAO,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AACnD,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AACvE,OAAO,EAAE,iBAAiB,EAAE,MAAM,iCAAiC,CAAC;AAEpE,OAAO,EAAE,iBAAiB,EAAE,MAAM,+BAA+B,CAAC;AAElE,OAAO,EAAE,aAAa,EAAE,MAAM,2BAA2B,CAAC;AAe1D;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CACjC,MAAc,EACd,QAAgB,EAChB,OAAe,EACf,MAAiB,EACjB,SAA0B,EAC1B,OAAO,GAAe,KAAK,EAC3B,YAA8B;IAE9B,MAAM,iBAAiB,GAAG,MAAM,CAAC,YAAY,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,iBAAiB,CAAC,WAAW,CAAC,CAAC;IAC/F,MAAM,QAAQ,GAAG,CAAC,IAAI,aAAa,CAAC,QAAQ,CAAC,EAAE,IAAI,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC;IAE1E,gEAAgE;IAChE,MAAM,QAAQ,GAAG,wBAAwB,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC1D,IAAI,QAAQ,EAAE,CAAC;QACb,kBAAkB,CAAC,QAAQ,EAAE,MAAM,CAAC,yBAAyB,CAAC,CAAC;IACjE,CAAC;IAED,6DAA6D;IAC7D,MAAM,MAAM,GAAG,IAAI,cAAc,CAAC,qBAAqB,EAAE,SAAS,EAAE,YAAY,CAAC,CAAC;IAClF,IAAI,SAAS,GAAG,IAAI,CAAC;IACrB,IAAI,YAAY,GAAG,EAAE,CAAC;IACtB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAC7B,IAAI,CAAC;QACH,MAAM,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,IAAI,WAAW,EAAE,CAAC,CAAC;QACtD,YAAY,GAAG,MAAM,MAAM,CAAC,eAAe,CAAC,QAAQ,CAAC,CAAC;IACxD,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,SAAS,GAAG,KAAK,CAAC;QAClB,YAAY,CAAC,yBAAyB,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAC5F,CAAC;YAAS,CAAC;QACT,MAAM,MAAM,CAAC,OAAO,EAAE,CAAC;IACzB,CAAC;IAED,wFAAwF;IACxF,wFAAwF;IACxF,wEAAwE;IACxE,IAAI,QAAQ,GAAgB,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC;IAC1C,IAAI,CAAC;QACH,MAAM,CAAC,GAAG,MAAM,CAAC,WAAW,EAAE,EAAE,CAAC;QACjC,IAAI,CAAC;YAAE,QAAQ,GAAG,CAAC,CAAC;IACtB,CAAC;IAAC,MAAM,CAAC;QACP,qDAAqD;IACvD,CAAC;IAED,iGAAiG;IACjG,sFAAsF;IACtF,gGAAgG;IAChG,iBAAiB,CAAC,MAAM,EAAE;QACxB,OAAO;QACP,OAAO,EAAE,aAAa,EAAE;QACxB,KAAK,EAAE,MAAM,CAAC,gBAAgB;QAC9B,MAAM,EAAE,OAAO;QACf,QAAQ,EAAE,YAAY;QACtB,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,YAAY,EAAE,QAAQ,CAAC,YAAY;QACnC,KAAK,EAAE,QAAQ,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS;QAC7D,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;KACnC,CAAC,CAAC;IAEH,iBAAiB,EAAE,IAAI,EAAE,CAAC;IAE1B,IAAI,MAAM,CAAC,iBAAiB,KAAK,KAAK,EAAE,CAAC;QACvC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,yEAAyE;IAC1F,CAAC;IACD,IAAI,QAAQ,EAAE,CAAC;QACb,IAAI,CAAC;YACH,eAAe,EAAE,CAAC;YAClB,kBAAkB,EAAE,CAAC;YACrB,cAAc,CAAC,mCAAmC,QAAQ,EAAE,CAAC,CAAC;QAChE,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,YAAY,CAAC,mCAAmC,QAAQ,EAAE,CAAC,CAAC;YAC5D,YAAY,CAAC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;QACvE,CAAC;IACH,CAAC;IAED,OAAO,EAAE,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,YAAY,EAAE,GAAG,QAAQ,EAAE,CAAC;AAC9D,CAAC"}
+{"version":3,"file":"singleShot.js","sourceRoot":"","sources":["../../src/runtime/singleShot.ts"],"names":[],"mappings":"AACA,OAAO,EACL,qBAAqB,EACrB,OAAO,EACP,YAAY,EACZ,cAAc,EACd,eAAe,EACf,kBAAkB,EAClB,kBAAkB,GACnB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AACnE,OAAO,EAAE,cAAc,EAAE,MAAM,6BAA6B,CAAC;AAC7D,OAAO,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AACnD,OAAO,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AACxD,OAAO,EAAE,iBAAiB,EAAE,MAAM,iCAAiC,CAAC;AAEpE,OAAO,EAAE,iBAAiB,EAAE,MAAM,+BAA+B,CAAC;AAElE,OAAO,EAAE,aAAa,EAAE,MAAM,2BAA2B,CAAC;AAe1D;;;;;;;;;;;;;;;;;;;;;GAqBG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CACjC,MAAc;AACd,2FAA2F;AAC3F,gGAAgG;AAChG,iGAAiG;AACjG,iGAAiG;AACjG,mGAAmG;AACnG,2FAA2F;AAC3F,iGAAiG;AACjG,gGAAgG;AAChG,oGAAoG;AACpG,SAAiB,EACjB,OAAe,EACf,MAAiB,EACjB,SAA0B,EAC1B,OAAO,GAAe,KAAK,EAC3B,YAA8B;IAE9B,MAAM,iBAAiB,GAAG,MAAM,CAAC,YAAY,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,iBAAiB,CAAC,WAAW,CAAC,CAAC;IAC/F,iGAAiG;IACjG,MAAM,QAAQ,GAAG,CAAC,IAAI,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC;IAE7C,gEAAgE;IAChE,MAAM,QAAQ,GAAG,wBAAwB,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC1D,IAAI,QAAQ,EAAE,CAAC;QACb,kBAAkB,CAAC,QAAQ,EAAE,MAAM,CAAC,yBAAyB,CAAC,CAAC;IACjE,CAAC;IAED,6DAA6D;IAC7D,MAAM,MAAM,GAAG,IAAI,cAAc,CAAC,qBAAqB,EAAE,SAAS,EAAE,YAAY,CAAC,CAAC;IAClF,IAAI,SAAS,GAAG,IAAI,CAAC;IACrB,IAAI,YAAY,GAAG,EAAE,CAAC;IACtB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAC7B,IAAI,CAAC;QACH,MAAM,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,MAAM,EAAE,IAAI,WAAW,EAAE,CAAC,CAAC;QACtD,YAAY,GAAG,MAAM,MAAM,CAAC,eAAe,CAAC,QAAQ,CAAC,CAAC;IACxD,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,SAAS,GAAG,KAAK,CAAC;QAClB,YAAY,CAAC,yBAAyB,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;IAC5F,CAAC;YAAS,CAAC;QACT,MAAM,MAAM,CAAC,OAAO,EAAE,CAAC;IACzB,CAAC;IAED,wFAAwF;IACxF,wFAAwF;IACxF,wEAAwE;IACxE,IAAI,QAAQ,GAAgB,EAAE,KAAK,EAAE,EAAE,EAAE,CAAC;IAC1C,IAAI,CAAC;QACH,MAAM,CAAC,GAAG,MAAM,CAAC,WAAW,EAAE,EAAE,CAAC;QACjC,IAAI,CAAC;YAAE,QAAQ,GAAG,CAAC,CAAC;IACtB,CAAC;IAAC,MAAM,CAAC;QACP,qDAAqD;IACvD,CAAC;IAED,iGAAiG;IACjG,sFAAsF;IACtF,gGAAgG;IAChG,iBAAiB,CAAC,MAAM,EAAE;QACxB,OAAO;QACP,OAAO,EAAE,aAAa,EAAE;QACxB,KAAK,EAAE,MAAM,CAAC,gBAAgB;QAC9B,MAAM,EAAE,OAAO;QACf,QAAQ,EAAE,YAAY;QACtB,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,YAAY,EAAE,QAAQ,CAAC,YAAY;QACnC,KAAK,EAAE,QAAQ,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS;QAC7D,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;KACnC,CAAC,CAAC;IAEH,iBAAiB,EAAE,IAAI,EAAE,CAAC;IAE1B,IAAI,MAAM,CAAC,iBAAiB,KAAK,KAAK,EAAE,CAAC;QACvC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,yEAAyE;IAC1F,CAAC;IACD,IAAI,QAAQ,EAAE,CAAC;QACb,IAAI,CAAC;YACH,eAAe,EAAE,CAAC;YAClB,kBAAkB,EAAE,CAAC;YACrB,cAAc,CAAC,mCAAmC,QAAQ,EAAE,CAAC,CAAC;QAChE,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,YAAY,CAAC,mCAAmC,QAAQ,EAAE,CAAC,CAAC;YAC5D,YAAY,CAAC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;QACvE,CAAC;IACH,CAAC;IAED,OAAO,EAAE,EAAE,EAAE,SAAS,EAAE,MAAM,EAAE,YAAY,EAAE,GAAG,QAAQ,EAAE,CAAC;AAC9D,CAAC"}

package/dist/utils/debugDump.d.ts

@@ -1,10 +1,11 @@
 /**
- * GS2-46 — `/debug-dump`: a live-session diagnostic archive, written UNSANITIZED (that's GS2-47's
- * job on top of this). The user typed the command themselves, mid-session, knowingly, so the
- * caller (the slash command) is responsible for the loud "may contain secrets" warning — this
- * module only writes the files.
+ * GS2-46/GS2-47 — `/debug-dump`: a live-session diagnostic archive. GS2-46 shipped it raw; GS2-47
+ * adds a shared secret-redaction pass ({@link file://./redactSecrets.ts}) that is ON BY DEFAULT and
+ * applied to EVERY artifact before it hits disk. The caller opts out via `redact: false`, in which
+ * case the archive is raw and the caller surfaces the loud "may contain secrets" warning; when
+ * redaction is on the caller shows a softened "secrets redacted; review before sharing" note.
  */
-/** Input to {@link writeDebugDump}. `transcript`/`config` are dumped as-is (raw, unsanitized). */
+/** Input to {@link writeDebugDump}. */
 export interface WriteDebugDumpInput {
     /** The full transcript (all turns, tool calls + results). Opaque — serialized as JSON. */
     transcript: unknown;
@@ -12,6 +13,13 @@ export interface WriteDebugDumpInput {
     config: unknown;
     /** Model display name, already resolved by the caller. */
     modelDisplayName?: string;
+    /**
+     * GS2-47 — apply the shared secret-redaction pass to every artifact before writing. Defaults to
+     * ON: omitted or any value other than the literal `false` redacts (read-site `!== false`, matching
+     * the config default so an opt-out has to be explicit). `false` writes a RAW archive (the caller
+     * is then responsible for the loud "may contain secrets" warning).
+     */
+    redact?: boolean;
     /**
      * Working directory for git-state collection. Defaults to `process.cwd()`; overridable for
      * tests so the "not a git repo" / "inside a git repo" paths are both exercisable without
@@ -23,13 +31,27 @@ export interface WriteDebugDumpResult {
     /** The absolute path to the archive directory just written. */
     archiveDir: string;
 }
+/**
+ * The filesystem-safe directory-name segment for one dump: the ISO timestamp with `:` and `.`
+ * (illegal on Windows, noisy everywhere) replaced by `-`. This is the ONLY path component
+ * debugDump generates — and the only one it is responsible for sanitizing. The parent it is joined
+ * under (the global `~/.gsloth` dir) is supplied by the environment and may legitimately carry a
+ * drive-letter colon on Windows (`C:\…`), which is not ours to strip. Exported so the invariant
+ * "the generated segment is colon-free" is testable on any platform without asserting anything
+ * about the (platform-dependent) parent path (GS2-50).
+ */
+export declare function debugDumpDirName(date?: Date): string;
 /**
  * Write one timestamped `/debug-dump` archive under the GLOBAL `~/.gsloth/debug-dumps/<timestamp>/`
  * (via `ensureGlobalGslothDir()` — mirrors how `resolveHistoryDbPath()` builds its path under the
  * same dir — NOT the per-project cwd-relative helper of the same name elsewhere in this codebase).
  * Contains: the full transcript, the resolved config, env/version info, the in-memory debugLog
- * ring buffer, and (best-effort) git repo state. Everything is dumped raw/unsanitized — this node
- * ships deliberately unsanitized (GS2-47 adds redaction on top later); the caller is responsible
- * for surfacing the "may contain secrets" warning to the user.
+ * ring buffer, and (best-effort) git repo state.
+ *
+ * GS2-47 — unless `input.redact === false`, the shared secret-redaction pass
+ * ({@link file://./redactSecrets.ts}) is applied to EVERY artifact before it is written: the literal
+ * values of secret-named env vars + inline config secrets are substituted everywhere, a tight set of
+ * provider-key/auth-header patterns is masked, and the config's sensitive fields are masked in place.
+ * On opt-out the archive is raw and the caller surfaces the loud "may contain secrets" warning.
  */
 export declare function writeDebugDump(input: WriteDebugDumpInput): WriteDebugDumpResult;