@xoxoai/checkmate 0.4.28 → 0.5.1

package/dist/tools/define-agent-tool.d.ts

@@ -8,37 +8,12 @@ type ToolConfig<TSchema extends z.ZodType> = {
     handler: (args: z.infer<TSchema>, context: AgentToolContext) => Promise<AgentToolResult> | AgentToolResult;
 };
 /**
- * Creates a typed Checkmate tool from a Zod schema and handler.
- *
- * @example
- * ```ts
- * import { defineAgentTool } from '@xoxoai/checkmate/core'
- * import { z } from 'zod/v4'
- *
- * const apiHealthTool = defineAgentTool({
- *   name: 'check_api_health',
- *   description: 'Check whether the API is healthy',
- *   schema: z.object({ url: z.string() }).strict(),
- *   handler: async ({ url }) => `API health is good for ${url}`,
- * })
- * ```
+ * Creates an internal runner tool, which may return a step assertion.
+ * Driver authors use defineDriverTool from '@xoxoai/checkmate/driver' instead.
  */
 export declare function defineAgentTool<TSchema extends z.ZodType>(toolConfig: ToolConfig<TSchema>): AgentTool;
 /**
- * Preferred alias for `defineAgentTool()`.
- *
- * @example
- * ```ts
- * import { defineTool } from '@xoxoai/checkmate/core'
- * import { z } from 'zod/v4'
- *
- * const tool = defineTool({
- *   name: 'check_api_health',
- *   description: 'Check whether the API is healthy',
- *   schema: z.object({ url: z.string() }).strict(),
- *   handler: async ({ url }) => `API health is good for ${url}`,
- * })
- * ```
+ * Internal shorthand for defineAgentTool; not part of the public driver API.
  */
 export declare const defineTool: typeof defineAgentTool;
 export {};

package/dist/tools/define-agent-tool.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"define-agent-tool.d.ts","sourceRoot":"","sources":["../../src/tools/define-agent-tool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,QAAQ,CAAA;AAC1B,OAAO,EAAE,SAAS,EAAE,gBAAgB,EAAuB,eAAe,EAAE,MAAM,YAAY,CAAA;AAE9F,KAAK,UAAU,CAAC,OAAO,SAAS,CAAC,CAAC,OAAO,IAAI;IAC5C,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,EAAE,MAAM,CAAA;IACnB,MAAM,EAAE,OAAO,CAAA;IACf,MAAM,CAAC,EAAE,OAAO,CAAA;IAChB,OAAO,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,gBAAgB,KAAK,OAAO,CAAC,eAAe,CAAC,GAAG,eAAe,CAAA;CAC1G,CAAA;AAED;;;;;;;;;;;;;;;GAeG;AACH,wBAAgB,eAAe,CAAC,OAAO,SAAS,CAAC,CAAC,OAAO,EAAE,UAAU,EAAE,UAAU,CAAC,OAAO,CAAC,GAAG,SAAS,CAsBrG;AAED;;;;;;;;;;;;;;;GAeG;AACH,eAAO,MAAM,UAAU,wBAAkB,CAAA"}
+{"version":3,"file":"define-agent-tool.d.ts","sourceRoot":"","sources":["../../src/tools/define-agent-tool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,QAAQ,CAAA;AAC1B,OAAO,EAAE,SAAS,EAAE,gBAAgB,EAAuB,eAAe,EAAE,MAAM,YAAY,CAAA;AAE9F,KAAK,UAAU,CAAC,OAAO,SAAS,CAAC,CAAC,OAAO,IAAI;IAC5C,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,EAAE,MAAM,CAAA;IACnB,MAAM,EAAE,OAAO,CAAA;IACf,MAAM,CAAC,EAAE,OAAO,CAAA;IAChB,OAAO,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,EAAE,OAAO,EAAE,gBAAgB,KAAK,OAAO,CAAC,eAAe,CAAC,GAAG,eAAe,CAAA;CAC1G,CAAA;AAED;;;GAGG;AACH,wBAAgB,eAAe,CAAC,OAAO,SAAS,CAAC,CAAC,OAAO,EAAE,UAAU,EAAE,UAAU,CAAC,OAAO,CAAC,GAAG,SAAS,CAsBrG;AAED;;GAEG;AACH,eAAO,MAAM,UAAU,wBAAkB,CAAA"}

package/dist/tools/define-agent-tool.js

@@ -1,19 +1,7 @@
 import { z } from 'zod/v4';
 /**
- * Creates a typed Checkmate tool from a Zod schema and handler.
- *
- * @example
- * ```ts
- * import { defineAgentTool } from '@xoxoai/checkmate/core'
- * import { z } from 'zod/v4'
- *
- * const apiHealthTool = defineAgentTool({
- *   name: 'check_api_health',
- *   description: 'Check whether the API is healthy',
- *   schema: z.object({ url: z.string() }).strict(),
- *   handler: async ({ url }) => `API health is good for ${url}`,
- * })
- * ```
+ * Creates an internal runner tool, which may return a step assertion.
+ * Driver authors use defineDriverTool from '@xoxoai/checkmate/driver' instead.
  */
 export function defineAgentTool(toolConfig) {
     const jsonSchema = z.toJSONSchema(toolConfig.schema);
@@ -36,20 +24,7 @@ export function defineAgentTool(toolConfig) {
     };
 }
 /**
- * Preferred alias for `defineAgentTool()`.
- *
- * @example
- * ```ts
- * import { defineTool } from '@xoxoai/checkmate/core'
- * import { z } from 'zod/v4'
- *
- * const tool = defineTool({
- *   name: 'check_api_health',
- *   description: 'Check whether the API is healthy',
- *   schema: z.object({ url: z.string() }).strict(),
- *   handler: async ({ url }) => `API health is good for ${url}`,
- * })
- * ```
+ * Internal shorthand for defineAgentTool; not part of the public driver API.
  */
 export const defineTool = defineAgentTool;
 //# sourceMappingURL=define-agent-tool.js.map

package/dist/tools/define-agent-tool.js.map

@@ -1 +1 @@
-{"version":3,"file":"define-agent-tool.js","sourceRoot":"","sources":["../../src/tools/define-agent-tool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,QAAQ,CAAA;AAW1B;;;;;;;;;;;;;;;GAeG;AACH,MAAM,UAAU,eAAe,CAA4B,UAA+B;IACzF,MAAM,UAAU,GAAG,CAAC,CAAC,YAAY,CAAC,UAAU,CAAC,MAAM,CAA4B,CAAA;IAC/E,OAAO,UAAU,CAAC,OAAO,CAAA;IAEzB,MAAM,UAAU,GAAwB;QACvC,IAAI,EAAE,UAAU,CAAC,IAAI;QACrB,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,UAAU,EAAE,UAAU;QACtB,MAAM,EAAE,UAAU,CAAC,MAAM,IAAI,IAAI;KACjC,CAAA;IAED,OAAO;QACN,UAAU;QACV,OAAO,EAAE,KAAK,EAAE,IAAa,EAAE,OAAyB,EAAE,EAAE;YAC3D,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,CAAA;YAChD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;gBACrB,OAAO,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,qBAAqB,UAAU,CAAC,IAAI,MAAM,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,EAAE,CAAC,CAAA;YACnG,CAAC;YAED,OAAO,UAAU,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,CAAA;QAChD,CAAC;KACD,CAAA;AACF,CAAC;AAED;;;;;;;;;;;;;;;GAeG;AACH,MAAM,CAAC,MAAM,UAAU,GAAG,eAAe,CAAA"}
+{"version":3,"file":"define-agent-tool.js","sourceRoot":"","sources":["../../src/tools/define-agent-tool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,QAAQ,CAAA;AAW1B;;;GAGG;AACH,MAAM,UAAU,eAAe,CAA4B,UAA+B;IACzF,MAAM,UAAU,GAAG,CAAC,CAAC,YAAY,CAAC,UAAU,CAAC,MAAM,CAA4B,CAAA;IAC/E,OAAO,UAAU,CAAC,OAAO,CAAA;IAEzB,MAAM,UAAU,GAAwB;QACvC,IAAI,EAAE,UAAU,CAAC,IAAI;QACrB,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,UAAU,EAAE,UAAU;QACtB,MAAM,EAAE,UAAU,CAAC,MAAM,IAAI,IAAI;KACjC,CAAA;IAED,OAAO;QACN,UAAU;QACV,OAAO,EAAE,KAAK,EAAE,IAAa,EAAE,OAAyB,EAAE,EAAE;YAC3D,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,CAAA;YAChD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;gBACrB,OAAO,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,qBAAqB,UAAU,CAAC,IAAI,MAAM,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,EAAE,CAAC,CAAA;YACnG,CAAC;YAED,OAAO,UAAU,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,EAAE,OAAO,CAAC,CAAA;QAChD,CAAC;KACD,CAAA;AACF,CAAC;AAED;;GAEG;AACH,MAAM,CAAC,MAAM,UAAU,GAAG,eAAe,CAAA"}

package/dist/tools/dispatcher.d.ts

@@ -1,11 +1,19 @@
-import { ToolRegistry, ToolResponse } from './registry.js';
-import { AgentToolContext, ToolCall } from './types.js';
+import type { RuntimeLogger } from '../logging/types.js';
+import { LoopDetector } from './loop-detector.js';
+import { ToolRegistry } from './registry.js';
+import { AgentToolContext, ToolCall, ToolResponse } from './types.js';
 export declare class ToolDispatcher {
     private readonly toolRegistry;
     private readonly loopDetector;
-    constructor(toolRegistry: ToolRegistry);
+    private readonly runtimeLogger;
+    private readonly debug;
+    constructor(toolRegistry: ToolRegistry, loopDetector: LoopDetector, runtimeLogger: RuntimeLogger, debug?: boolean);
     getToolRegistry(): ToolRegistry;
     dispatch(toolCall: ToolCall, context: AgentToolContext): Promise<ToolResponse | null>;
+    private formatAllowedToolNames;
     private normalizeToolResponse;
 }
+export declare class ToolDispatchError extends Error {
+    constructor(message: string, options?: ErrorOptions);
+}
 //# sourceMappingURL=dispatcher.d.ts.map

package/dist/tools/dispatcher.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"dispatcher.d.ts","sourceRoot":"","sources":["../../src/tools/dispatcher.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,YAAY,EAAE,YAAY,EAAE,MAAM,eAAe,CAAA;AAC1D,OAAO,EAAE,gBAAgB,EAAmB,QAAQ,EAAE,MAAM,YAAY,CAAA;AAExE,qBAAa,cAAc;IAGd,OAAO,CAAC,QAAQ,CAAC,YAAY;IAFzC,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAc;gBAEd,YAAY,EAAE,YAAY;IAIvD,eAAe,IAAI,YAAY;IAIzB,QAAQ,CAAC,QAAQ,EAAE,QAAQ,EAAE,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAAC,YAAY,GAAG,IAAI,CAAC;IAa3F,OAAO,CAAC,qBAAqB;CAqB7B"}
+{"version":3,"file":"dispatcher.d.ts","sourceRoot":"","sources":["../../src/tools/dispatcher.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAA;AAExD,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAA;AACjD,OAAO,EAAE,YAAY,EAAE,MAAM,eAAe,CAAA;AAC5C,OAAO,EAAE,gBAAgB,EAAsC,QAAQ,EAAE,YAAY,EAAE,MAAM,YAAY,CAAA;AAEzG,qBAAa,cAAc;IAEzB,OAAO,CAAC,QAAQ,CAAC,YAAY;IAC7B,OAAO,CAAC,QAAQ,CAAC,YAAY;IAC7B,OAAO,CAAC,QAAQ,CAAC,aAAa;IAC9B,OAAO,CAAC,QAAQ,CAAC,KAAK;gBAHL,YAAY,EAAE,YAAY,EAC1B,YAAY,EAAE,YAAY,EAC1B,aAAa,EAAE,aAAa,EAC5B,KAAK,UAAQ;IAG/B,eAAe,IAAI,YAAY;IAIzB,QAAQ,CAAC,QAAQ,EAAE,QAAQ,EAAE,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAAC,YAAY,GAAG,IAAI,CAAC;IA6C3F,OAAO,CAAC,sBAAsB;IAK9B,OAAO,CAAC,qBAAqB;CAkB7B;AAED,qBAAa,iBAAkB,SAAQ,KAAK;gBAC/B,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,YAAY;CAInD"}

package/dist/tools/dispatcher.js

@@ -1,24 +1,54 @@
-import { logger } from '../logging/index.js';
-import { LoopDetector } from './loop-detector.js';
+import { scrub } from '../redaction/scrub.js';
 export class ToolDispatcher {
     toolRegistry;
     loopDetector;
-    constructor(toolRegistry) {
+    runtimeLogger;
+    debug;
+    constructor(toolRegistry, loopDetector, runtimeLogger, debug = false) {
         this.toolRegistry = toolRegistry;
-        this.loopDetector = new LoopDetector(toolRegistry.getRuntimeConfig().getLoopMaxRepetitions());
+        this.loopDetector = loopDetector;
+        this.runtimeLogger = runtimeLogger;
+        this.debug = debug;
     }
     getToolRegistry() {
         return this.toolRegistry;
     }
     async dispatch(toolCall, context) {
         this.loopDetector.recordToolCall(toolCall);
-        logger.info(`executing tool: ${toolCall.name}:\n${JSON.stringify(toolCall.arguments ?? {}, null, 2)}`);
+        this.runtimeLogger.info(`executing tool: ${toolCall.name}:\n${JSON.stringify(toolCall.arguments ?? {}, null, 2)}`);
         const tool = this.toolRegistry.resolve(toolCall.name);
         if (!tool) {
-            throw new Error(`Invalid tool name, received call\n: ${JSON.stringify(toolCall, null, 2)}`);
+            throw new ToolDispatchError([
+                `Invalid tool name: ${toolCall.name}`,
+                `arguments: ${preview(toolCall.arguments, 1_000)}`,
+                `registered_tools: ${this.toolRegistry.getRegisteredToolNames().join(', ') || '(none)'}`,
+                `allowed_tools: ${this.formatAllowedToolNames()}`,
+            ].join('\n'));
         }
-        const result = await tool.execute(toolCall.arguments, context);
-        return this.normalizeToolResponse(toolCall.name, result);
+        let result;
+        try {
+            result = await tool.execute(toolCall.arguments, context);
+        }
+        catch (error) {
+            throw new ToolDispatchError([
+                `Tool execution failed: ${toolCall.name}`,
+                `arguments: ${preview(toolCall.arguments, 1_000)}`,
+                `original_error: ${error instanceof Error ? error.message : String(error)}`,
+            ].join('\n'), { cause: error });
+        }
+        const response = this.normalizeToolResponse(toolCall.name, result);
+        if (response === null && this.debug) {
+            this.runtimeLogger.debug([
+                'tool completed without model response:',
+                `tool: ${toolCall.name}`,
+                `arguments: ${preview(toolCall.arguments ?? {}, 1_000)}`,
+            ].join('\n'));
+        }
+        return response;
+    }
+    formatAllowedToolNames() {
+        const allowedNames = this.toolRegistry.getAllowedToolNames();
+        return allowedNames === '*' ? '(all registered tools allowed)' : allowedNames.join(', ') || '(none)';
     }
     normalizeToolResponse(toolName, result) {
         if (result === undefined) {
@@ -33,9 +63,21 @@ export class ToolDispatcher {
             response: normalizedResult.response,
             snapshot: normalizedResult.snapshot ?? null,
             status: normalizedResult.status ?? 'success',
+            ...(normalizedResult.assertion ? { assertion: normalizedResult.assertion } : {}),
         };
     }
 }
+export class ToolDispatchError extends Error {
+    constructor(message, options) {
+        super(message, options);
+        this.name = 'ToolDispatchError';
+    }
+}
+function preview(value, maxLength) {
+    const text = typeof value === 'string' ? value : JSON.stringify(value, null, 2);
+    const safeText = scrub(text ?? String(value));
+    return safeText.length <= maxLength ? safeText : `${safeText.slice(0, maxLength - 3)}...`;
+}
 function inferToolResponseStatus(response) {
     const normalizedResponse = response.trim().toLowerCase();
     if (normalizedResponse.startsWith('failed') ||

package/dist/tools/dispatcher.js.map

@@ -1 +1 @@
-{"version":3,"file":"dispatcher.js","sourceRoot":"","sources":["../../src/tools/dispatcher.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,qBAAqB,CAAA;AAC5C,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAA;AAIjD,MAAM,OAAO,cAAc;IAGG;IAFZ,YAAY,CAAc;IAE3C,YAA6B,YAA0B;QAA1B,iBAAY,GAAZ,YAAY,CAAc;QACtD,IAAI,CAAC,YAAY,GAAG,IAAI,YAAY,CAAC,YAAY,CAAC,gBAAgB,EAAE,CAAC,qBAAqB,EAAE,CAAC,CAAA;IAC9F,CAAC;IAED,eAAe;QACd,OAAO,IAAI,CAAC,YAAY,CAAA;IACzB,CAAC;IAED,KAAK,CAAC,QAAQ,CAAC,QAAkB,EAAE,OAAyB;QAC3D,IAAI,CAAC,YAAY,CAAC,cAAc,CAAC,QAAQ,CAAC,CAAA;QAC1C,MAAM,CAAC,IAAI,CAAC,mBAAmB,QAAQ,CAAC,IAAI,MAAM,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,IAAI,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC,CAAA;QAEtG,MAAM,IAAI,GAAG,IAAI,CAAC,YAAY,CAAC,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAA;QACrD,IAAI,CAAC,IAAI,EAAE,CAAC;YACX,MAAM,IAAI,KAAK,CAAC,uCAAuC,IAAI,CAAC,SAAS,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC,CAAA;QAC5F,CAAC;QAED,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC,CAAA;QAC9D,OAAO,IAAI,CAAC,qBAAqB,CAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAA;IACzD,CAAC;IAEO,qBAAqB,CAAC,QAAgB,EAAE,MAAuB;QACtE,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;YAC1B,OAAO,IAAI,CAAA;QACZ,CAAC;QAED,IAAI,OAAO,MAAM,KAAK,QAAQ,EAAE,CAAC;YAChC,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,uBAAuB,CAAC,MAAM,CAAC,EAAE,CAAA;QACrF,CAAC;QAED,MAAM,gBAAgB,GAAG,CAAC,MAAM,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,CAInD,CAAA;QACD,OAAO;YACN,IAAI,EAAE,QAAQ;YACd,QAAQ,EAAE,gBAAgB,CAAC,QAAQ;YACnC,QAAQ,EAAE,gBAAgB,CAAC,QAAQ,IAAI,IAAI;YAC3C,MAAM,EAAE,gBAAgB,CAAC,MAAM,IAAI,SAAS;SAC5C,CAAA;IACF,CAAC;CACD;AAED,SAAS,uBAAuB,CAAC,QAAgB;IAChD,MAAM,kBAAkB,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAA;IACxD,IACC,kBAAkB,CAAC,UAAU,CAAC,QAAQ,CAAC;QACvC,kBAAkB,CAAC,UAAU,CAAC,OAAO,CAAC;QACtC,kBAAkB,CAAC,UAAU,CAAC,iBAAiB,CAAC;QAChD,kBAAkB,CAAC,UAAU,CAAC,UAAU,CAAC,EACxC,CAAC;QACF,OAAO,OAAO,CAAA;IACf,CAAC;IAED,OAAO,SAAS,CAAA;AACjB,CAAC"}
+{"version":3,"file":"dispatcher.js","sourceRoot":"","sources":["../../src/tools/dispatcher.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAA;AAK7C,MAAM,OAAO,cAAc;IAER;IACA;IACA;IACA;IAJlB,YACkB,YAA0B,EAC1B,YAA0B,EAC1B,aAA4B,EAC5B,QAAQ,KAAK;QAHb,iBAAY,GAAZ,YAAY,CAAc;QAC1B,iBAAY,GAAZ,YAAY,CAAc;QAC1B,kBAAa,GAAb,aAAa,CAAe;QAC5B,UAAK,GAAL,KAAK,CAAQ;IAC5B,CAAC;IAEJ,eAAe;QACd,OAAO,IAAI,CAAC,YAAY,CAAA;IACzB,CAAC;IAED,KAAK,CAAC,QAAQ,CAAC,QAAkB,EAAE,OAAyB;QAC3D,IAAI,CAAC,YAAY,CAAC,cAAc,CAAC,QAAQ,CAAC,CAAA;QAC1C,IAAI,CAAC,aAAa,CAAC,IAAI,CACtB,mBAAmB,QAAQ,CAAC,IAAI,MAAM,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,IAAI,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,CACzF,CAAA;QAED,MAAM,IAAI,GAAG,IAAI,CAAC,YAAY,CAAC,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAA;QACrD,IAAI,CAAC,IAAI,EAAE,CAAC;YACX,MAAM,IAAI,iBAAiB,CAC1B;gBACC,sBAAsB,QAAQ,CAAC,IAAI,EAAE;gBACrC,cAAc,OAAO,CAAC,QAAQ,CAAC,SAAS,EAAE,KAAK,CAAC,EAAE;gBAClD,qBAAqB,IAAI,CAAC,YAAY,CAAC,sBAAsB,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,QAAQ,EAAE;gBACxF,kBAAkB,IAAI,CAAC,sBAAsB,EAAE,EAAE;aACjD,CAAC,IAAI,CAAC,IAAI,CAAC,CACZ,CAAA;QACF,CAAC;QAED,IAAI,MAAuB,CAAA;QAC3B,IAAI,CAAC;YACJ,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC,CAAA;QACzD,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,MAAM,IAAI,iBAAiB,CAC1B;gBACC,0BAA0B,QAAQ,CAAC,IAAI,EAAE;gBACzC,cAAc,OAAO,CAAC,QAAQ,CAAC,SAAS,EAAE,KAAK,CAAC,EAAE;gBAClD,mBAAmB,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE;aAC3E,CAAC,IAAI,CAAC,IAAI,CAAC,EACZ,EAAE,KAAK,EAAE,KAAK,EAAE,CAChB,CAAA;QACF,CAAC;QAED,MAAM,QAAQ,GAAG,IAAI,CAAC,qBAAqB,CAAC,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAA;QAClE,IAAI,QAAQ,KAAK,IAAI,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;YACrC,IAAI,CAAC,aAAa,CAAC,KAAK,CACvB;gBACC,wCAAwC;gBACxC,SAAS,QAAQ,CAAC,IAAI,EAAE;gBACxB,cAAc,OAAO,CAAC,QAAQ,CAAC,SAAS,IAAI,EAAE,EAAE,KAAK,CAAC,EAAE;aACxD,CAAC,IAAI,CAAC,IAAI,CAAC,CACZ,CAAA;QACF,CAAC;QACD,OAAO,QAAQ,CAAA;IAChB,CAAC;IAEO,sBAAsB;QAC7B,MAAM,YAAY,GAAG,IAAI,CAAC,YAAY,CAAC,mBAAmB,EAAE,CAAA;QAC5D,OAAO,YAAY,KAAK,GAAG,CAAC,CAAC,CAAC,gCAAgC,CAAC,CAAC,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,QAAQ,CAAA;IACrG,CAAC;IAEO,qBAAqB,CAAC,QAAgB,EAAE,MAAuB;QACtE,IAAI,MAAM,KAAK,SAAS,EAAE,CAAC;YAC1B,OAAO,IAAI,CAAA;QACZ,CAAC;QAED,IAAI,OAAO,MAAM,KAAK,QAAQ,EAAE,CAAC;YAChC,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,uBAAuB,CAAC,MAAM,CAAC,EAAE,CAAA;QACrF,CAAC;QAED,MAAM,gBAAgB,GAAG,CAAC,MAAM,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAsB,CAAA;QAC1E,OAAO;YACN,IAAI,EAAE,QAAQ;YACd,QAAQ,EAAE,gBAAgB,CAAC,QAAQ;YACnC,QAAQ,EAAE,gBAAgB,CAAC,QAAQ,IAAI,IAAI;YAC3C,MAAM,EAAE,gBAAgB,CAAC,MAAM,IAAI,SAAS;YAC5C,GAAG,CAAC,gBAAgB,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,gBAAgB,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAChF,CAAA;IACF,CAAC;CACD;AAED,MAAM,OAAO,iBAAkB,SAAQ,KAAK;IAC3C,YAAY,OAAe,EAAE,OAAsB;QAClD,KAAK,CAAC,OAAO,EAAE,OAAO,CAAC,CAAA;QACvB,IAAI,CAAC,IAAI,GAAG,mBAAmB,CAAA;IAChC,CAAC;CACD;AAED,SAAS,OAAO,CAAC,KAAc,EAAE,SAAiB;IACjD,MAAM,IAAI,GAAG,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,CAAA;IAC/E,MAAM,QAAQ,GAAG,KAAK,CAAC,IAAI,IAAI,MAAM,CAAC,KAAK,CAAC,CAAC,CAAA;IAC7C,OAAO,QAAQ,CAAC,MAAM,IAAI,SAAS,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,GAAG,CAAC,CAAC,KAAK,CAAA;AAC1F,CAAC;AAED,SAAS,uBAAuB,CAAC,QAAgB;IAChD,MAAM,kBAAkB,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAA;IACxD,IACC,kBAAkB,CAAC,UAAU,CAAC,QAAQ,CAAC;QACvC,kBAAkB,CAAC,UAAU,CAAC,OAAO,CAAC;QACtC,kBAAkB,CAAC,UAAU,CAAC,iBAAiB,CAAC;QAChD,kBAAkB,CAAC,UAAU,CAAC,UAAU,CAAC,EACxC,CAAC;QACF,OAAO,OAAO,CAAA;IACf,CAAC;IAED,OAAO,SAAS,CAAA;AACjB,CAAC"}

package/dist/tools/registry.d.ts

@@ -1,21 +1,19 @@
 import { ChatCompletionFunctionTool } from 'openai/resources/chat/completions';
-import { RuntimeConfig } from '../config/runtime-config.js';
 import { AgentTool } from './types.js';
-export type ToolResponse = {
-    name?: string;
-    response: string;
-    snapshot?: string | null;
-    status: 'success' | 'error';
-};
+export type { ToolResponse } from './types.js';
 export declare class ToolRegistry {
-    private readonly runtimeConfig;
     private readonly tools;
     private readonly toolsByName;
-    constructor(runtimeConfig: RuntimeConfig);
+    private readonly allowedNames;
+    constructor(input: {
+        allowedTools: '*' | readonly string[];
+    });
     register(tool: AgentTool | AgentTool[]): void;
-    getRuntimeConfig(): RuntimeConfig;
     resolve(toolName: string): AgentTool | undefined;
+    getRegisteredToolNames(): string[];
     getTools(): Promise<ChatCompletionFunctionTool[]>;
+    getAllowedToolNames(): '*' | readonly string[];
+    private isAllowed;
     private toOpenAiTool;
 }
 //# sourceMappingURL=registry.d.ts.map

package/dist/tools/registry.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"registry.d.ts","sourceRoot":"","sources":["../../src/tools/registry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,0BAA0B,EAAE,MAAM,mCAAmC,CAAA;AAC9E,OAAO,EAAE,aAAa,EAAE,MAAM,6BAA6B,CAAA;AAC3D,OAAO,EAAE,SAAS,EAAe,MAAM,YAAY,CAAA;AAEnD,MAAM,MAAM,YAAY,GAAG;IAC1B,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,MAAM,CAAA;IAChB,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IACxB,MAAM,EAAE,SAAS,GAAG,OAAO,CAAA;CAC3B,CAAA;AAED,qBAAa,YAAY;IAIZ,OAAO,CAAC,QAAQ,CAAC,aAAa;IAH1C,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAkB;IACxC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAA+B;gBAE9B,aAAa,EAAE,aAAa;IAEzD,QAAQ,CAAC,IAAI,EAAE,SAAS,GAAG,SAAS,EAAE,GAAG,IAAI;IAc7C,gBAAgB,IAAI,aAAa;IAIjC,OAAO,CAAC,QAAQ,EAAE,MAAM,GAAG,SAAS,GAAG,SAAS;IAI1C,QAAQ,IAAI,OAAO,CAAC,0BAA0B,EAAE,CAAC;IAWvD,OAAO,CAAC,YAAY;CAWpB"}
+{"version":3,"file":"registry.d.ts","sourceRoot":"","sources":["../../src/tools/registry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,0BAA0B,EAAE,MAAM,mCAAmC,CAAA;AAE9E,OAAO,EAAE,SAAS,EAAe,MAAM,YAAY,CAAA;AAEnD,YAAY,EAAE,YAAY,EAAE,MAAM,YAAY,CAAA;AAE9C,qBAAa,YAAY;IACxB,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAkB;IACxC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAA+B;IAC3D,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAyB;gBAE1C,KAAK,EAAE;QAAE,YAAY,EAAE,GAAG,GAAG,SAAS,MAAM,EAAE,CAAA;KAAE;IAI5D,QAAQ,CAAC,IAAI,EAAE,SAAS,GAAG,SAAS,EAAE,GAAG,IAAI;IAc7C,OAAO,CAAC,QAAQ,EAAE,MAAM,GAAG,SAAS,GAAG,SAAS;IAKhD,sBAAsB,IAAI,MAAM,EAAE;IAI5B,QAAQ,IAAI,OAAO,CAAC,0BAA0B,EAAE,CAAC;IAMvD,mBAAmB,IAAI,GAAG,GAAG,SAAS,MAAM,EAAE;IAI9C,OAAO,CAAC,SAAS;IASjB,OAAO,CAAC,YAAY;CAWpB"}

package/dist/tools/registry.js

@@ -1,10 +1,11 @@
+import { StepResultTool } from './step/result-tool.js';
 import { getToolName } from './types.js';
 export class ToolRegistry {
-    runtimeConfig;
     tools = [];
     toolsByName = new Map();
-    constructor(runtimeConfig) {
-        this.runtimeConfig = runtimeConfig;
+    allowedNames;
+    constructor(input) {
+        this.allowedNames = input.allowedTools === '*' ? '*' : [...input.allowedTools];
     }
     register(tool) {
         const tools = Array.isArray(tool) ? tool : [tool];
@@ -17,19 +18,26 @@ export class ToolRegistry {
             this.toolsByName.set(toolName, registeredTool);
         }
     }
-    getRuntimeConfig() {
-        return this.runtimeConfig;
-    }
     resolve(toolName) {
+        if (!this.isAllowed(toolName))
+            return undefined;
         return this.toolsByName.get(toolName);
     }
+    getRegisteredToolNames() {
+        return this.tools.map((tool) => getToolName(tool));
+    }
     async getTools() {
-        const allowedNames = this.runtimeConfig.getAllowedFunctionNames();
         const definitions = this.tools.map((tool) => this.toOpenAiTool(tool));
-        if (allowedNames.length === 0) {
-            return definitions;
-        }
-        return definitions.filter((tool) => allowedNames.includes(tool.function.name));
+        return definitions.filter((tool) => this.isAllowed(tool.function.name));
+    }
+    getAllowedToolNames() {
+        return this.allowedNames === '*' ? '*' : [...this.allowedNames];
+    }
+    isAllowed(toolName) {
+        return (toolName === StepResultTool.TOOL_FAIL_TEST_STEP ||
+            toolName === StepResultTool.TOOL_PASS_TEST_STEP ||
+            this.allowedNames === '*' ||
+            this.allowedNames.includes(toolName));
     }
     toOpenAiTool(tool) {
         return {

package/dist/tools/registry.js.map

@@ -1 +1 @@
-{"version":3,"file":"registry.js","sourceRoot":"","sources":["../../src/tools/registry.ts"],"names":[],"mappings":"AAEA,OAAO,EAAa,WAAW,EAAE,MAAM,YAAY,CAAA;AASnD,MAAM,OAAO,YAAY;IAIK;IAHZ,KAAK,GAAgB,EAAE,CAAA;IACvB,WAAW,GAAG,IAAI,GAAG,EAAqB,CAAA;IAE3D,YAA6B,aAA4B;QAA5B,kBAAa,GAAb,aAAa,CAAe;IAAG,CAAC;IAE7D,QAAQ,CAAC,IAA6B;QACrC,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAA;QAEjD,KAAK,MAAM,cAAc,IAAI,KAAK,EAAE,CAAC;YACpC,MAAM,QAAQ,GAAG,WAAW,CAAC,cAAc,CAAC,CAAA;YAC5C,IAAI,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;gBACpC,MAAM,IAAI,KAAK,CAAC,oCAAoC,QAAQ,GAAG,CAAC,CAAA;YACjE,CAAC;YAED,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,cAAc,CAAC,CAAA;YAC/B,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,QAAQ,EAAE,cAAc,CAAC,CAAA;QAC/C,CAAC;IACF,CAAC;IAED,gBAAgB;QACf,OAAO,IAAI,CAAC,aAAa,CAAA;IAC1B,CAAC;IAED,OAAO,CAAC,QAAgB;QACvB,OAAO,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAA;IACtC,CAAC;IAED,KAAK,CAAC,QAAQ;QACb,MAAM,YAAY,GAAG,IAAI,CAAC,aAAa,CAAC,uBAAuB,EAAE,CAAA;QACjE,MAAM,WAAW,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,CAAA;QAErE,IAAI,YAAY,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC/B,OAAO,WAAW,CAAA;QACnB,CAAC;QAED,OAAO,WAAW,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,YAAY,CAAC,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAA;IAC/E,CAAC;IAEO,YAAY,CAAC,IAAe;QACnC,OAAO;YACN,IAAI,EAAE,UAAU;YAChB,QAAQ,EAAE;gBACT,IAAI,EAAE,IAAI,CAAC,UAAU,CAAC,IAAI;gBAC1B,WAAW,EAAE,IAAI,CAAC,UAAU,CAAC,WAAW;gBACxC,UAAU,EAAE,IAAI,CAAC,UAAU,CAAC,UAAU;gBACtC,MAAM,EAAE,IAAI,CAAC,UAAU,CAAC,MAAM;aAC9B;SACD,CAAA;IACF,CAAC;CACD"}
+{"version":3,"file":"registry.js","sourceRoot":"","sources":["../../src/tools/registry.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAA;AACtD,OAAO,EAAa,WAAW,EAAE,MAAM,YAAY,CAAA;AAInD,MAAM,OAAO,YAAY;IACP,KAAK,GAAgB,EAAE,CAAA;IACvB,WAAW,GAAG,IAAI,GAAG,EAAqB,CAAA;IAC1C,YAAY,CAAyB;IAEtD,YAAY,KAAgD;QAC3D,IAAI,CAAC,YAAY,GAAG,KAAK,CAAC,YAAY,KAAK,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,YAAY,CAAC,CAAA;IAC/E,CAAC;IAED,QAAQ,CAAC,IAA6B;QACrC,MAAM,KAAK,GAAG,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAA;QAEjD,KAAK,MAAM,cAAc,IAAI,KAAK,EAAE,CAAC;YACpC,MAAM,QAAQ,GAAG,WAAW,CAAC,cAAc,CAAC,CAAA;YAC5C,IAAI,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;gBACpC,MAAM,IAAI,KAAK,CAAC,oCAAoC,QAAQ,GAAG,CAAC,CAAA;YACjE,CAAC;YAED,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,cAAc,CAAC,CAAA;YAC/B,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,QAAQ,EAAE,cAAc,CAAC,CAAA;QAC/C,CAAC;IACF,CAAC;IAED,OAAO,CAAC,QAAgB;QACvB,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC;YAAE,OAAO,SAAS,CAAA;QAC/C,OAAO,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAA;IACtC,CAAC;IAED,sBAAsB;QACrB,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAA;IACnD,CAAC;IAED,KAAK,CAAC,QAAQ;QACb,MAAM,WAAW,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,CAAA;QAErE,OAAO,WAAW,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAA;IACxE,CAAC;IAED,mBAAmB;QAClB,OAAO,IAAI,CAAC,YAAY,KAAK,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,YAAY,CAAC,CAAA;IAChE,CAAC;IAEO,SAAS,CAAC,QAAgB;QACjC,OAAO,CACN,QAAQ,KAAK,cAAc,CAAC,mBAAmB;YAC/C,QAAQ,KAAK,cAAc,CAAC,mBAAmB;YAC/C,IAAI,CAAC,YAAY,KAAK,GAAG;YACzB,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,QAAQ,CAAC,CACpC,CAAA;IACF,CAAC;IAEO,YAAY,CAAC,IAAe;QACnC,OAAO;YACN,IAAI,EAAE,UAAU;YAChB,QAAQ,EAAE;gBACT,IAAI,EAAE,IAAI,CAAC,UAAU,CAAC,IAAI;gBAC1B,WAAW,EAAE,IAAI,CAAC,UAAU,CAAC,WAAW;gBACxC,UAAU,EAAE,IAAI,CAAC,UAAU,CAAC,UAAU;gBACtC,MAAM,EAAE,IAAI,CAAC,UAAU,CAAC,MAAM;aAC9B;SACD,CAAA;IACF,CAAC;CACD"}

package/dist/tools/step/result-tool.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"result-tool.d.ts","sourceRoot":"","sources":["../../../src/tools/step/result-tool.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,SAAS,EAAE,MAAM,aAAa,CAAA;AAEvC,eAAO,MAAM,cAAc;;;CAGjB,CAAA;AAQV,wBAAgB,qBAAqB,IAAI,SAAS,EAAE,CAmBnD"}
+{"version":3,"file":"result-tool.d.ts","sourceRoot":"","sources":["../../../src/tools/step/result-tool.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,SAAS,EAAE,MAAM,aAAa,CAAA;AAEvC,eAAO,MAAM,cAAc;;;CAGjB,CAAA;AAQV,wBAAgB,qBAAqB,IAAI,SAAS,EAAE,CAqBnD"}

package/dist/tools/step/result-tool.js

@@ -15,17 +15,19 @@ export function createStepResultTools() {
             name: StepResultTool.TOOL_FAIL_TEST_STEP,
             description: 'Fail the test step with the actual result',
             schema: stepResultSchema,
-            handler: ({ actualResult }, context) => {
-                context.resolveStepResult({ passed: false, actual: actualResult });
-            },
+            handler: ({ actualResult }) => ({
+                response: actualResult,
+                assertion: { passed: false, actual: actualResult },
+            }),
         }),
         defineAgentTool({
             name: StepResultTool.TOOL_PASS_TEST_STEP,
             description: 'Pass the test step with the actual result',
             schema: stepResultSchema,
-            handler: ({ actualResult }, context) => {
-                context.resolveStepResult({ passed: true, actual: actualResult });
-            },
+            handler: ({ actualResult }) => ({
+                response: actualResult,
+                assertion: { passed: true, actual: actualResult },
+            }),
         }),
     ];
 }

package/dist/tools/step/result-tool.js.map

@@ -1 +1 @@
-{"version":3,"file":"result-tool.js","sourceRoot":"","sources":["../../../src/tools/step/result-tool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,QAAQ,CAAA;AAC1B,OAAO,EAAE,eAAe,EAAE,MAAM,yBAAyB,CAAA;AAGzD,MAAM,CAAC,MAAM,cAAc,GAAG;IAC7B,mBAAmB,EAAE,gBAAgB;IACrC,mBAAmB,EAAE,gBAAgB;CAC5B,CAAA;AAEV,MAAM,gBAAgB,GAAG,CAAC;KACxB,MAAM,CAAC;IACP,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,oCAAoC,CAAC;CACvE,CAAC;KACD,MAAM,EAAE,CAAA;AAEV,MAAM,UAAU,qBAAqB;IACpC,OAAO;QACN,eAAe,CAAC;YACf,IAAI,EAAE,cAAc,CAAC,mBAAmB;YACxC,WAAW,EAAE,2CAA2C;YACxD,MAAM,EAAE,gBAAgB;YACxB,OAAO,EAAE,CAAC,EAAE,YAAY,EAAE,EAAE,OAAO,EAAE,EAAE;gBACtC,OAAO,CAAC,iBAAiB,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,YAAY,EAAE,CAAC,CAAA;YACnE,CAAC;SACD,CAAC;QACF,eAAe,CAAC;YACf,IAAI,EAAE,cAAc,CAAC,mBAAmB;YACxC,WAAW,EAAE,2CAA2C;YACxD,MAAM,EAAE,gBAAgB;YACxB,OAAO,EAAE,CAAC,EAAE,YAAY,EAAE,EAAE,OAAO,EAAE,EAAE;gBACtC,OAAO,CAAC,iBAAiB,CAAC,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,YAAY,EAAE,CAAC,CAAA;YAClE,CAAC;SACD,CAAC;KACF,CAAA;AACF,CAAC"}
+{"version":3,"file":"result-tool.js","sourceRoot":"","sources":["../../../src/tools/step/result-tool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,QAAQ,CAAA;AAC1B,OAAO,EAAE,eAAe,EAAE,MAAM,yBAAyB,CAAA;AAGzD,MAAM,CAAC,MAAM,cAAc,GAAG;IAC7B,mBAAmB,EAAE,gBAAgB;IACrC,mBAAmB,EAAE,gBAAgB;CAC5B,CAAA;AAEV,MAAM,gBAAgB,GAAG,CAAC;KACxB,MAAM,CAAC;IACP,YAAY,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,oCAAoC,CAAC;CACvE,CAAC;KACD,MAAM,EAAE,CAAA;AAEV,MAAM,UAAU,qBAAqB;IACpC,OAAO;QACN,eAAe,CAAC;YACf,IAAI,EAAE,cAAc,CAAC,mBAAmB;YACxC,WAAW,EAAE,2CAA2C;YACxD,MAAM,EAAE,gBAAgB;YACxB,OAAO,EAAE,CAAC,EAAE,YAAY,EAAE,EAAE,EAAE,CAAC,CAAC;gBAC/B,QAAQ,EAAE,YAAY;gBACtB,SAAS,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,YAAY,EAAE;aAClD,CAAC;SACF,CAAC;QACF,eAAe,CAAC;YACf,IAAI,EAAE,cAAc,CAAC,mBAAmB;YACxC,WAAW,EAAE,2CAA2C;YACxD,MAAM,EAAE,gBAAgB;YACxB,OAAO,EAAE,CAAC,EAAE,YAAY,EAAE,EAAE,EAAE,CAAC,CAAC;gBAC/B,QAAQ,EAAE,YAAY;gBACtB,SAAS,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,YAAY,EAAE;aACjD,CAAC;SACF,CAAC;KACF,CAAA;AACF,CAAC"}

package/dist/tools/tool-contract.d.ts

@@ -1,3 +1,3 @@
-export type { AgentTool, AgentToolContext, AgentToolDefinition, AgentToolResponse, AgentToolResult, ToolCall, } from './types.js';
+export type { AgentTool, AgentToolContext, AgentToolDefinition, AgentToolResponse, AgentToolResult, ToolCall, ToolExecution, ToolResponse, } from './types.js';
 export { getToolName } from './types.js';
 //# sourceMappingURL=tool-contract.d.ts.map

package/dist/tools/tool-contract.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tool-contract.d.ts","sourceRoot":"","sources":["../../src/tools/tool-contract.ts"],"names":[],"mappings":"AAAA,YAAY,EACX,SAAS,EACT,gBAAgB,EAChB,mBAAmB,EACnB,iBAAiB,EACjB,eAAe,EACf,QAAQ,GACR,MAAM,YAAY,CAAA;AACnB,OAAO,EAAE,WAAW,EAAE,MAAM,YAAY,CAAA"}
+{"version":3,"file":"tool-contract.d.ts","sourceRoot":"","sources":["../../src/tools/tool-contract.ts"],"names":[],"mappings":"AAAA,YAAY,EACX,SAAS,EACT,gBAAgB,EAChB,mBAAmB,EACnB,iBAAiB,EACjB,eAAe,EACf,QAAQ,EACR,aAAa,EACb,YAAY,GACZ,MAAM,YAAY,CAAA;AACnB,OAAO,EAAE,WAAW,EAAE,MAAM,YAAY,CAAA"}

package/dist/tools/tool-contract.js.map

@@ -1 +1 @@
-{"version":3,"file":"tool-contract.js","sourceRoot":"","sources":["../../src/tools/tool-contract.ts"],"names":[],"mappings":"AAQA,OAAO,EAAE,WAAW,EAAE,MAAM,YAAY,CAAA"}
+{"version":3,"file":"tool-contract.js","sourceRoot":"","sources":["../../src/tools/tool-contract.ts"],"names":[],"mappings":"AAUA,OAAO,EAAE,WAAW,EAAE,MAAM,YAAY,CAAA"}

package/dist/tools/types.d.ts

@@ -1,4 +1,5 @@
-import { ResolveStepResult, Step } from '../runtime/types.js';
+import { Step, StepAssertion } from '../runtime/types.js';
+import type { StepControl } from '../runtime/scenario-control.js';
 /**
  * Normalized tool call emitted by the model.
  *
@@ -75,6 +76,10 @@ export type AgentToolResponse = {
      * Outcome status used in tool execution summaries.
      */
     status?: 'success' | 'error';
+    /**
+     * Ends the step with this assertion instead of running another model turn.
+     */
+    assertion?: StepAssertion;
 };
 /**
  * Any valid value a tool can return.
@@ -96,7 +101,7 @@ export type AgentToolResult = AgentToolResponse | string | void;
  * @example
  * ```ts
  * const handler = async (_args: unknown, context: AgentToolContext) => {
- *   context.resolveStepResult({ passed: true, actual: context.step.expect })
+ *   return `Working on: ${context.step.action}`
  * }
  * ```
  */
@@ -106,14 +111,16 @@ export type AgentToolContext = {
      */
     step: Step;
     /**
-     * Callback used to finish the step.
+     * Model turn the tool is being dispatched on, starting at `1`, when known.
      */
-    resolveStepResult: ResolveStepResult;
+    turn?: number;
+    signal?: AbortSignal;
+    control?: StepControl;
 };
 /**
- * Tool contract used by the runner loop.
+ * Internal tool contract used by the runner loop, including assertion-bearing tools.
  *
- * Most users should create tools with `defineTool()` instead of building this object manually.
+ * Driver authors use DriverTool and defineDriverTool from '@xoxoai/checkmate/driver'.
  *
  * @example
  * ```ts
@@ -126,6 +133,7 @@ export type AgentToolContext = {
  * ```
  */
 export type AgentTool = {
+    driverId?: string;
     /**
      * Tool definition exposed to the model.
      */
@@ -135,5 +143,68 @@ export type AgentTool = {
      */
     execute: (args: unknown, context: AgentToolContext) => Promise<AgentToolResult> | AgentToolResult;
 };
+/**
+ * Normalized tool response recorded by the runner loop.
+ *
+ * @example
+ * ```ts
+ * const response: ToolResponse = {
+ *   name: 'browser_navigate',
+ *   response: 'Navigated to https://example.com',
+ *   snapshot: null,
+ *   status: 'success',
+ * }
+ * ```
+ */
+export type ToolResponse = {
+    /**
+     * Tool that produced the response.
+     */
+    name?: string;
+    /**
+     * Human-readable result returned to the model.
+     */
+    response: string;
+    /**
+     * Fresh page snapshot captured after the tool ran.
+     */
+    snapshot?: string | null;
+    /**
+     * Outcome status used in tool execution summaries.
+     */
+    status: 'success' | 'error';
+    /**
+     * Assertion that ends the step, when the tool produced one.
+     */
+    assertion?: StepAssertion;
+};
+/**
+ * One tool execution observed during the runner loop.
+ *
+ * Driver hooks receive the public DriverToolExecution shape instead.
+ *
+ * @example
+ * ```ts
+ * const execution: ToolExecution = {
+ *   toolCallId: 'call_1',
+ *   toolCall: { name: 'browser_navigate', arguments: { url: 'https://example.com' } },
+ *   toolResponse: { response: 'Navigated to example.com', status: 'success' },
+ * }
+ * ```
+ */
+export type ToolExecution = {
+    /**
+     * Provider-issued id for the tool call.
+     */
+    toolCallId: string;
+    /**
+     * Normalized tool call request.
+     */
+    toolCall: ToolCall;
+    /**
+     * Normalized tool response returned by Checkmate.
+     */
+    toolResponse: ToolResponse;
+};
 export declare function getToolName(tool: AgentTool): string;
 //# sourceMappingURL=types.d.ts.map

package/dist/tools/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/tools/types.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,IAAI,EAAE,MAAM,qBAAqB,CAAA;AAE7D;;;;;;;;;;GAUG;AACH,MAAM,MAAM,QAAQ,GAAG;IACtB;;OAEG;IACH,IAAI,EAAE,MAAM,CAAA;IAEZ;;OAEG;IACH,SAAS,CAAC,EAAE,OAAO,CAAA;CACnB,CAAA;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,MAAM,mBAAmB,GAAG;IACjC;;OAEG;IACH,IAAI,EAAE,MAAM,CAAA;IAEZ;;OAEG;IACH,WAAW,EAAE,MAAM,CAAA;IAEnB;;OAEG;IACH,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAEnC;;OAEG;IACH,MAAM,EAAE,OAAO,CAAA;CACf,CAAA;AAED;;;;;;;;;;GAUG;AACH,MAAM,MAAM,iBAAiB,GAAG;IAC/B;;OAEG;IACH,QAAQ,EAAE,MAAM,CAAA;IAEhB;;OAEG;IACH,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IAExB;;OAEG;IACH,MAAM,CAAC,EAAE,SAAS,GAAG,OAAO,CAAA;CAC5B,CAAA;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,MAAM,eAAe,GAAG,iBAAiB,GAAG,MAAM,GAAG,IAAI,CAAA;AAE/D;;;;;;;;;GASG;AACH,MAAM,MAAM,gBAAgB,GAAG;IAC9B;;OAEG;IACH,IAAI,EAAE,IAAI,CAAA;IAEV;;OAEG;IACH,iBAAiB,EAAE,iBAAiB,CAAA;CACpC,CAAA;AAED;;;;;;;;;;;;;;GAcG;AACH,MAAM,MAAM,SAAS,GAAG;IACvB;;OAEG;IACH,UAAU,EAAE,mBAAmB,CAAA;IAE/B;;OAEG;IACH,OAAO,EAAE,CAAC,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,gBAAgB,KAAK,OAAO,CAAC,eAAe,CAAC,GAAG,eAAe,CAAA;CACjG,CAAA;AAED,wBAAgB,WAAW,CAAC,IAAI,EAAE,SAAS,GAAG,MAAM,CAEnD"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/tools/types.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAA;AACzD,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,gCAAgC,CAAA;AAEjE;;;;;;;;;;GAUG;AACH,MAAM,MAAM,QAAQ,GAAG;IACtB;;OAEG;IACH,IAAI,EAAE,MAAM,CAAA;IAEZ;;OAEG;IACH,SAAS,CAAC,EAAE,OAAO,CAAA;CACnB,CAAA;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,MAAM,mBAAmB,GAAG;IACjC;;OAEG;IACH,IAAI,EAAE,MAAM,CAAA;IAEZ;;OAEG;IACH,WAAW,EAAE,MAAM,CAAA;IAEnB;;OAEG;IACH,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;IAEnC;;OAEG;IACH,MAAM,EAAE,OAAO,CAAA;CACf,CAAA;AAED;;;;;;;;;;GAUG;AACH,MAAM,MAAM,iBAAiB,GAAG;IAC/B;;OAEG;IACH,QAAQ,EAAE,MAAM,CAAA;IAEhB;;OAEG;IACH,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IAExB;;OAEG;IACH,MAAM,CAAC,EAAE,SAAS,GAAG,OAAO,CAAA;IAE5B;;OAEG;IACH,SAAS,CAAC,EAAE,aAAa,CAAA;CACzB,CAAA;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,MAAM,eAAe,GAAG,iBAAiB,GAAG,MAAM,GAAG,IAAI,CAAA;AAE/D;;;;;;;;;GASG;AACH,MAAM,MAAM,gBAAgB,GAAG;IAC9B;;OAEG;IACH,IAAI,EAAE,IAAI,CAAA;IAEV;;OAEG;IACH,IAAI,CAAC,EAAE,MAAM,CAAA;IAEb,MAAM,CAAC,EAAE,WAAW,CAAA;IAEpB,OAAO,CAAC,EAAE,WAAW,CAAA;CACrB,CAAA;AAED;;;;;;;;;;;;;;GAcG;AACH,MAAM,MAAM,SAAS,GAAG;IACvB,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB;;OAEG;IACH,UAAU,EAAE,mBAAmB,CAAA;IAE/B;;OAEG;IACH,OAAO,EAAE,CAAC,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,gBAAgB,KAAK,OAAO,CAAC,eAAe,CAAC,GAAG,eAAe,CAAA;CACjG,CAAA;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,MAAM,YAAY,GAAG;IAC1B;;OAEG;IACH,IAAI,CAAC,EAAE,MAAM,CAAA;IAEb;;OAEG;IACH,QAAQ,EAAE,MAAM,CAAA;IAEhB;;OAEG;IACH,QAAQ,CAAC,EAAE,MAAM,GAAG,IAAI,CAAA;IAExB;;OAEG;IACH,MAAM,EAAE,SAAS,GAAG,OAAO,CAAA;IAE3B;;OAEG;IACH,SAAS,CAAC,EAAE,aAAa,CAAA;CACzB,CAAA;AAED;;;;;;;;;;;;;GAaG;AACH,MAAM,MAAM,aAAa,GAAG;IAC3B;;OAEG;IACH,UAAU,EAAE,MAAM,CAAA;IAElB;;OAEG;IACH,QAAQ,EAAE,QAAQ,CAAA;IAElB;;OAEG;IACH,YAAY,EAAE,YAAY,CAAA;CAC1B,CAAA;AAED,wBAAgB,WAAW,CAAC,IAAI,EAAE,SAAS,GAAG,MAAM,CAEnD"}

package/dist/tools/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/tools/types.ts"],"names":[],"mappings":"AAwJA,MAAM,UAAU,WAAW,CAAC,IAAe;IAC1C,OAAO,IAAI,CAAC,UAAU,CAAC,IAAI,CAAA;AAC5B,CAAC"}
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../../src/tools/types.ts"],"names":[],"mappings":"AA0OA,MAAM,UAAU,WAAW,CAAC,IAAe;IAC1C,OAAO,IAAI,CAAC,UAAU,CAAC,IAAI,CAAA;AAC5B,CAAC"}

package/docs/CLI.md

@@ -0,0 +1,59 @@
+# CLI Reference
+
+The Checkmate CLI is non-interactive. Machine-readable commands write exactly one complete JSON document to stdout. Diagnostics and discarded-worker-stream notices use stderr.
+
+## Commands
+
+```text
+checkmate run <request.json|-> [--config <path>]
+checkmate validate <request.json|-> [--config <path>]
+checkmate describe [--config <path>]
+checkmate --help
+checkmate --version
+```
+
+`-` reads the request from stdin. Request file paths and explicit config paths are resolved from the current working directory. Without `--config`, Checkmate reads exactly `$PWD/checkmate.config.json`; it does not walk parent directories.
+
+## Run
+
+`run` performs bounded preparation in the CLI parent. Invalid input returns an `InvalidInvocationResultV1`. An operational preparation failure or identity-allocation failure returns a parent-authored `PreExecutionOperationalResultV1` with `infra / pre-execution-error`, exit `3`, `targetMutation: "not-attempted"`, and no scenario, step, usage, or evidence claims. A valid request reserves one run directory and starts a child worker. The worker executes only the prepared plan; it does not reread configuration, reprobe preflight secrets, or allocate another identity.
+
+The parent is the sole stdout writer. It accepts worker output only after validating the complete IPC frame, digest, deterministic bytes, `RunResultV1` schema, immutable prepared identity, policy, limits, step grammar, usage, and committed evidence references. Successful committed output is byte-identical to the run's `result.json`.
+
+If no authoritative worker result is accepted before the applicable cutoff, the parent emits `ContainmentResultV1`. Containment has no semantic `steps`, usage, evidence-completeness claim, or committed result reference.
+
+## Validate
+
+`validate` performs the same static preparation as `run`, including secret availability, but stops before identity allocation, executable import, driver startup, target mutation, or output creation.
+
+## Describe
+
+`describe` returns schema references, policy metadata, registered static driver descriptors, logical secret requirements, and reason-specific routing guidance. It neither imports drivers nor probes secret values.
+
+## Exit Codes
+
+| Code | Route                                                                                 |
+| ---: | ------------------------------------------------------------------------------------- |
+|  `0` | Passed run, valid request, or available description                                   |
+|  `1` | Application expectation failure                                                       |
+|  `2` | Model failure                                                                         |
+|  `3` | Infrastructure failure, interruption, containment, or unexpected static-command error |
+|  `4` | Invalid invocation or unavailable requested configuration                             |
+
+Exit codes provide coarse shell routing. The stdout document remains the detailed contract.
+
+## Signals And Cutoffs
+
+Scenario timing starts when the parent forks the worker, so driver import and startup consume the scenario limit. Normal `cleanup-started` disarms the scenario timer and starts one fresh cleanup cutoff. A scenario timeout or first `SIGINT`/`SIGTERM` sends one cooperative abort and immediately starts one cleanup/termination bound; a later cleanup notification cannot extend it.
+
+The parent compares its monotonic clock with the absolute active deadline before every state transition and after asynchronous terminal reconciliation. After an external signal it accepts only `interrupted`, `driver-teardown-failed`, or `result-write-failed`. After a run deadline it accepts only `scenario-timeout`, `driver-teardown-failed`, or `result-write-failed`. Unrelated pass, application, and model results cannot override a latched stop cause.
+
+A worker-finalized interrupted result accepted in time exits `3`. A second signal after an external stop forces the first signal's native route: `130` for `SIGINT` or `143` for `SIGTERM`, with no fabricated stdout envelope.
+
+## Output Discipline
+
+- Worker stdout and stderr bytes are counted, drained, and discarded.
+- Worker logger diagnostics are exact-secret and pattern sanitized before IPC.
+- The parent pattern-sanitizes diagnostic messages again before stderr.
+- Diagnostic `code` and RFC 6901 `path` fields remain structural and are not rewritten.
+- Terminal result bytes are never reserialized or modified by the parent.