@probelabs/probe 0.6.0-rc211 → 0.6.0-rc213

package/build/agent/ProbeAgent.js

@@ -2832,6 +2832,10 @@ Follow these instructions carefully:
           if (this.enableDelegate && this.allowedTools.isEnabled('delegate')) {
             validTools.push('delegate');
           }
+          // Analyze All tool (for bulk data processing with map-reduce)
+          if (this.allowedTools.isEnabled('analyze_all')) {
+            validTools.push('analyze_all');
+          }
           // Task tool (require both enableTasks flag AND allowedTools permission)
           if (this.enableTasks && this.allowedTools.isEnabled('task')) {
             validTools.push('task');

package/build/agent/index.js

@@ -4250,28 +4250,44 @@ ${chunkSummaries}`;
 }
 async function planAnalysis(question, path9, options) {
   if (options.debug) {
-    console.error(`[analyze_all] Phase 1: Planning analysis strategy...`);
+    console.error(`[analyze_all] Phase 1: Planning analysis strategy (with exploration)...`);
   }
-  const planningTask = `Create an analysis plan for this question about a codebase:
+  const planningTask = `You are planning a bulk data analysis. Your goal is to find the BEST search strategy through exploration and experimentation.
 
-"${question}"
+QUESTION TO ANSWER: "${question}"
+SEARCH SCOPE: ${path9}
+
+YOUR TASK: Explore, experiment, then plan.
+
+STEP 1: EXPLORE THE REPOSITORY STRUCTURE
+Use listFiles to understand:
+- What types of files exist? (code, markdown, configs, etc.)
+- What are the naming patterns? (e.g., "Playbook", "Debrief", "spec", "test")
+- What directories and subdirectories exist?
+
+STEP 2: TEST DIFFERENT SEARCH QUERIES
+Run several experimental searches to find what works:
+- Try different keyword combinations related to the question
+- Test queries based on file naming patterns you discovered
+- Check which queries return relevant results vs empty results
+- Iterate until you find queries that actually return data
+
+For example, if looking for customer data:
+- First try: search for "customer" - see what comes back
+- If empty, try: search for "playbook" or "debrief" (common doc names)
+- Check the actual content to understand terminology used
 
-Search scope: ${path9}
+STEP 3: CREATE THE FINAL PLAN
+Based on your experiments, output the BEST search strategy.
 
-Use attempt_completion to output your plan in this EXACT format:
+Use attempt_completion with this EXACT format:
 
-SEARCH_QUERY: <elasticsearch query using OR for multiple terms, quotes for exact phrases>
+SEARCH_QUERY: <the query that WORKED in your experiments - use OR for multiple terms>
 AGGREGATION: <summarize | list_unique | count | group_by>
 EXTRACTION_PROMPT: <what to extract from each search result>
-REASONING: <brief explanation>
+REASONING: <what you discovered, what queries you tested, why the final query works>
 
-Example plan:
-SEARCH_QUERY: export OR function OR class OR tool
-AGGREGATION: list_unique
-EXTRACTION_PROMPT: Extract tool names and their purpose
-REASONING: Using list_unique to deduplicate tool definitions
-
-IMPORTANT: Use attempt_completion immediately with your plan. Do NOT try to search or answer the question - just create the analysis plan.`;
+CRITICAL: Do NOT guess keywords. Actually run searches and see what returns results!`;
   try {
     const result = await delegate({
       task: planningTask,
@@ -4284,10 +4300,9 @@ IMPORTANT: Use attempt_completion immediately with your plan. Do NOT try to sear
       tracer: options.tracer,
       enableBash: false,
       promptType: "code-researcher",
-      allowedTools: [],
-      // attempt_completion only (default tool)
-      maxIterations: 3,
-      timeout: 60
+      // Full tool access for exploration and experimentation
+      maxIterations: 15,
+      timeout: 180
     });
     const plan = parsePlanningResult(stripResultTags(result));
     if (options.debug) {
@@ -9610,7 +9625,7 @@ var init_common = __esm({
       env: external_exports.record(external_exports.string()).optional().describe("Additional environment variables (optional)")
     });
     analyzeAllSchema = external_exports.object({
-      question: external_exports.string().min(1).describe('Free-form question to answer (e.g., "What features are customers using?", "List all API endpoints"). The AI will automatically plan the search strategy, process all matching data, and synthesize a comprehensive answer.'),
+      question: external_exports.string().min(1).describe('Free-form question to answer (e.g., "What features are customers using?", "List all API endpoints"). The AI will automatically explore the repository, test search strategies, and synthesize a comprehensive answer.'),
       path: external_exports.string().optional().default(".").describe("Directory path to search in")
     });
     attemptCompletionSchema = {
@@ -71965,6 +71980,9 @@ You are working with a repository located at: ${searchDirectory}
               if (this.enableDelegate && this.allowedTools.isEnabled("delegate")) {
                 validTools.push("delegate");
               }
+              if (this.allowedTools.isEnabled("analyze_all")) {
+                validTools.push("analyze_all");
+              }
               if (this.enableTasks && this.allowedTools.isEnabled("task")) {
                 validTools.push("task");
               }

package/build/tools/analyzeAll.js

@@ -339,6 +339,8 @@ Organize all findings into clear categories with items listed under each.${compl
 
 /**
  * PHASE 1: Planning - Analyze the question and determine optimal strategy
+ * Uses a full agent that can explore the repository to understand its structure
+ * before creating the search plan.
  * @param {string} question - The user's free-form question
  * @param {string} path - Path to search in
  * @param {Object} options - Delegate options
@@ -346,32 +348,48 @@ Organize all findings into clear categories with items listed under each.${compl
  */
 async function planAnalysis(question, path, options) {
 	if (options.debug) {
-		console.error(`[analyze_all] Phase 1: Planning analysis strategy...`);
+		console.error(`[analyze_all] Phase 1: Planning analysis strategy (with exploration)...`);
 	}
 
-	const planningTask = `Create an analysis plan for this question about a codebase:
+	const planningTask = `You are planning a bulk data analysis. Your goal is to find the BEST search strategy through exploration and experimentation.
 
-"${question}"
+QUESTION TO ANSWER: "${question}"
+SEARCH SCOPE: ${path}
+
+YOUR TASK: Explore, experiment, then plan.
+
+STEP 1: EXPLORE THE REPOSITORY STRUCTURE
+Use listFiles to understand:
+- What types of files exist? (code, markdown, configs, etc.)
+- What are the naming patterns? (e.g., "Playbook", "Debrief", "spec", "test")
+- What directories and subdirectories exist?
+
+STEP 2: TEST DIFFERENT SEARCH QUERIES
+Run several experimental searches to find what works:
+- Try different keyword combinations related to the question
+- Test queries based on file naming patterns you discovered
+- Check which queries return relevant results vs empty results
+- Iterate until you find queries that actually return data
 
-Search scope: ${path}
+For example, if looking for customer data:
+- First try: search for "customer" - see what comes back
+- If empty, try: search for "playbook" or "debrief" (common doc names)
+- Check the actual content to understand terminology used
 
-Use attempt_completion to output your plan in this EXACT format:
+STEP 3: CREATE THE FINAL PLAN
+Based on your experiments, output the BEST search strategy.
 
-SEARCH_QUERY: <elasticsearch query using OR for multiple terms, quotes for exact phrases>
+Use attempt_completion with this EXACT format:
+
+SEARCH_QUERY: <the query that WORKED in your experiments - use OR for multiple terms>
 AGGREGATION: <summarize | list_unique | count | group_by>
 EXTRACTION_PROMPT: <what to extract from each search result>
-REASONING: <brief explanation>
-
-Example plan:
-SEARCH_QUERY: export OR function OR class OR tool
-AGGREGATION: list_unique
-EXTRACTION_PROMPT: Extract tool names and their purpose
-REASONING: Using list_unique to deduplicate tool definitions
+REASONING: <what you discovered, what queries you tested, why the final query works>
 
-IMPORTANT: Use attempt_completion immediately with your plan. Do NOT try to search or answer the question - just create the analysis plan.`;
+CRITICAL: Do NOT guess keywords. Actually run searches and see what returns results!`;
 
 	try {
-		// Planning phase - attempt_completion only
+		// Planning phase - full agent with exploration capabilities
 		const result = await delegate({
 			task: planningTask,
 			debug: options.debug,
@@ -383,9 +401,9 @@ IMPORTANT: Use attempt_completion immediately with your plan. Do NOT try to sear
 			tracer: options.tracer,
 			enableBash: false,
 			promptType: 'code-researcher',
-			allowedTools: [], // attempt_completion only (default tool)
-			maxIterations: 3,
-			timeout: 60
+			// Full tool access for exploration and experimentation
+			maxIterations: 15,
+			timeout: 180
 		});
 
 		const plan = parsePlanningResult(stripResultTags(result));

package/build/tools/common.js

@@ -47,7 +47,7 @@ export const bashSchema = z.object({
 });
 
 export const analyzeAllSchema = z.object({
-	question: z.string().min(1).describe('Free-form question to answer (e.g., "What features are customers using?", "List all API endpoints"). The AI will automatically plan the search strategy, process all matching data, and synthesize a comprehensive answer.'),
+	question: z.string().min(1).describe('Free-form question to answer (e.g., "What features are customers using?", "List all API endpoints"). The AI will automatically explore the repository, test search strategies, and synthesize a comprehensive answer.'),
 	path: z.string().optional().default('.').describe('Directory path to search in')
 });
 

package/cjs/agent/ProbeAgent.cjs

@@ -33576,28 +33576,44 @@ ${chunkSummaries}`;
 }
 async function planAnalysis(question, path9, options) {
   if (options.debug) {
-    console.error(`[analyze_all] Phase 1: Planning analysis strategy...`);
+    console.error(`[analyze_all] Phase 1: Planning analysis strategy (with exploration)...`);
   }
-  const planningTask = `Create an analysis plan for this question about a codebase:
+  const planningTask = `You are planning a bulk data analysis. Your goal is to find the BEST search strategy through exploration and experimentation.
 
-"${question}"
+QUESTION TO ANSWER: "${question}"
+SEARCH SCOPE: ${path9}
+
+YOUR TASK: Explore, experiment, then plan.
+
+STEP 1: EXPLORE THE REPOSITORY STRUCTURE
+Use listFiles to understand:
+- What types of files exist? (code, markdown, configs, etc.)
+- What are the naming patterns? (e.g., "Playbook", "Debrief", "spec", "test")
+- What directories and subdirectories exist?
+
+STEP 2: TEST DIFFERENT SEARCH QUERIES
+Run several experimental searches to find what works:
+- Try different keyword combinations related to the question
+- Test queries based on file naming patterns you discovered
+- Check which queries return relevant results vs empty results
+- Iterate until you find queries that actually return data
+
+For example, if looking for customer data:
+- First try: search for "customer" - see what comes back
+- If empty, try: search for "playbook" or "debrief" (common doc names)
+- Check the actual content to understand terminology used
 
-Search scope: ${path9}
+STEP 3: CREATE THE FINAL PLAN
+Based on your experiments, output the BEST search strategy.
 
-Use attempt_completion to output your plan in this EXACT format:
+Use attempt_completion with this EXACT format:
 
-SEARCH_QUERY: <elasticsearch query using OR for multiple terms, quotes for exact phrases>
+SEARCH_QUERY: <the query that WORKED in your experiments - use OR for multiple terms>
 AGGREGATION: <summarize | list_unique | count | group_by>
 EXTRACTION_PROMPT: <what to extract from each search result>
-REASONING: <brief explanation>
+REASONING: <what you discovered, what queries you tested, why the final query works>
 
-Example plan:
-SEARCH_QUERY: export OR function OR class OR tool
-AGGREGATION: list_unique
-EXTRACTION_PROMPT: Extract tool names and their purpose
-REASONING: Using list_unique to deduplicate tool definitions
-
-IMPORTANT: Use attempt_completion immediately with your plan. Do NOT try to search or answer the question - just create the analysis plan.`;
+CRITICAL: Do NOT guess keywords. Actually run searches and see what returns results!`;
   try {
     const result = await delegate({
       task: planningTask,
@@ -33610,10 +33626,9 @@ IMPORTANT: Use attempt_completion immediately with your plan. Do NOT try to sear
       tracer: options.tracer,
       enableBash: false,
       promptType: "code-researcher",
-      allowedTools: [],
-      // attempt_completion only (default tool)
-      maxIterations: 3,
-      timeout: 60
+      // Full tool access for exploration and experimentation
+      maxIterations: 15,
+      timeout: 180
     });
     const plan = parsePlanningResult(stripResultTags(result));
     if (options.debug) {
@@ -38936,7 +38951,7 @@ var init_common2 = __esm({
       env: external_exports.record(external_exports.string()).optional().describe("Additional environment variables (optional)")
     });
     analyzeAllSchema = external_exports.object({
-      question: external_exports.string().min(1).describe('Free-form question to answer (e.g., "What features are customers using?", "List all API endpoints"). The AI will automatically plan the search strategy, process all matching data, and synthesize a comprehensive answer.'),
+      question: external_exports.string().min(1).describe('Free-form question to answer (e.g., "What features are customers using?", "List all API endpoints"). The AI will automatically explore the repository, test search strategies, and synthesize a comprehensive answer.'),
       path: external_exports.string().optional().default(".").describe("Directory path to search in")
     });
     attemptCompletionSchema = {
@@ -101034,6 +101049,9 @@ You are working with a repository located at: ${searchDirectory}
               if (this.enableDelegate && this.allowedTools.isEnabled("delegate")) {
                 validTools.push("delegate");
               }
+              if (this.allowedTools.isEnabled("analyze_all")) {
+                validTools.push("analyze_all");
+              }
               if (this.enableTasks && this.allowedTools.isEnabled("task")) {
                 validTools.push("task");
               }

package/cjs/index.cjs

@@ -38267,7 +38267,7 @@ var init_common2 = __esm({
       env: external_exports.record(external_exports.string()).optional().describe("Additional environment variables (optional)")
     });
     analyzeAllSchema = external_exports.object({
-      question: external_exports.string().min(1).describe('Free-form question to answer (e.g., "What features are customers using?", "List all API endpoints"). The AI will automatically plan the search strategy, process all matching data, and synthesize a comprehensive answer.'),
+      question: external_exports.string().min(1).describe('Free-form question to answer (e.g., "What features are customers using?", "List all API endpoints"). The AI will automatically explore the repository, test search strategies, and synthesize a comprehensive answer.'),
       path: external_exports.string().optional().default(".").describe("Directory path to search in")
     });
     attemptCompletionSchema = {
@@ -97945,6 +97945,9 @@ You are working with a repository located at: ${searchDirectory}
               if (this.enableDelegate && this.allowedTools.isEnabled("delegate")) {
                 validTools.push("delegate");
               }
+              if (this.allowedTools.isEnabled("analyze_all")) {
+                validTools.push("analyze_all");
+              }
               if (this.enableTasks && this.allowedTools.isEnabled("task")) {
                 validTools.push("task");
               }
@@ -99649,28 +99652,44 @@ ${chunkSummaries}`;
 }
 async function planAnalysis(question, path9, options) {
   if (options.debug) {
-    console.error(`[analyze_all] Phase 1: Planning analysis strategy...`);
+    console.error(`[analyze_all] Phase 1: Planning analysis strategy (with exploration)...`);
   }
-  const planningTask = `Create an analysis plan for this question about a codebase:
+  const planningTask = `You are planning a bulk data analysis. Your goal is to find the BEST search strategy through exploration and experimentation.
 
-"${question}"
+QUESTION TO ANSWER: "${question}"
+SEARCH SCOPE: ${path9}
+
+YOUR TASK: Explore, experiment, then plan.
+
+STEP 1: EXPLORE THE REPOSITORY STRUCTURE
+Use listFiles to understand:
+- What types of files exist? (code, markdown, configs, etc.)
+- What are the naming patterns? (e.g., "Playbook", "Debrief", "spec", "test")
+- What directories and subdirectories exist?
+
+STEP 2: TEST DIFFERENT SEARCH QUERIES
+Run several experimental searches to find what works:
+- Try different keyword combinations related to the question
+- Test queries based on file naming patterns you discovered
+- Check which queries return relevant results vs empty results
+- Iterate until you find queries that actually return data
 
-Search scope: ${path9}
+For example, if looking for customer data:
+- First try: search for "customer" - see what comes back
+- If empty, try: search for "playbook" or "debrief" (common doc names)
+- Check the actual content to understand terminology used
 
-Use attempt_completion to output your plan in this EXACT format:
+STEP 3: CREATE THE FINAL PLAN
+Based on your experiments, output the BEST search strategy.
 
-SEARCH_QUERY: <elasticsearch query using OR for multiple terms, quotes for exact phrases>
+Use attempt_completion with this EXACT format:
+
+SEARCH_QUERY: <the query that WORKED in your experiments - use OR for multiple terms>
 AGGREGATION: <summarize | list_unique | count | group_by>
 EXTRACTION_PROMPT: <what to extract from each search result>
-REASONING: <brief explanation>
-
-Example plan:
-SEARCH_QUERY: export OR function OR class OR tool
-AGGREGATION: list_unique
-EXTRACTION_PROMPT: Extract tool names and their purpose
-REASONING: Using list_unique to deduplicate tool definitions
+REASONING: <what you discovered, what queries you tested, why the final query works>
 
-IMPORTANT: Use attempt_completion immediately with your plan. Do NOT try to search or answer the question - just create the analysis plan.`;
+CRITICAL: Do NOT guess keywords. Actually run searches and see what returns results!`;
   try {
     const result = await delegate({
       task: planningTask,
@@ -99683,10 +99702,9 @@ IMPORTANT: Use attempt_completion immediately with your plan. Do NOT try to sear
       tracer: options.tracer,
       enableBash: false,
       promptType: "code-researcher",
-      allowedTools: [],
-      // attempt_completion only (default tool)
-      maxIterations: 3,
-      timeout: 60
+      // Full tool access for exploration and experimentation
+      maxIterations: 15,
+      timeout: 180
     });
     const plan = parsePlanningResult(stripResultTags(result));
     if (options.debug) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@probelabs/probe",
-  "version": "0.6.0-rc211",
+  "version": "0.6.0-rc213",
   "description": "Node.js wrapper for the probe code search tool",
   "main": "src/index.js",
   "module": "src/index.js",

package/src/agent/ProbeAgent.js

@@ -2832,6 +2832,10 @@ Follow these instructions carefully:
           if (this.enableDelegate && this.allowedTools.isEnabled('delegate')) {
             validTools.push('delegate');
           }
+          // Analyze All tool (for bulk data processing with map-reduce)
+          if (this.allowedTools.isEnabled('analyze_all')) {
+            validTools.push('analyze_all');
+          }
           // Task tool (require both enableTasks flag AND allowedTools permission)
           if (this.enableTasks && this.allowedTools.isEnabled('task')) {
             validTools.push('task');

package/src/tools/analyzeAll.js

@@ -339,6 +339,8 @@ Organize all findings into clear categories with items listed under each.${compl
 
 /**
  * PHASE 1: Planning - Analyze the question and determine optimal strategy
+ * Uses a full agent that can explore the repository to understand its structure
+ * before creating the search plan.
  * @param {string} question - The user's free-form question
  * @param {string} path - Path to search in
  * @param {Object} options - Delegate options
@@ -346,32 +348,48 @@ Organize all findings into clear categories with items listed under each.${compl
  */
 async function planAnalysis(question, path, options) {
 	if (options.debug) {
-		console.error(`[analyze_all] Phase 1: Planning analysis strategy...`);
+		console.error(`[analyze_all] Phase 1: Planning analysis strategy (with exploration)...`);
 	}
 
-	const planningTask = `Create an analysis plan for this question about a codebase:
+	const planningTask = `You are planning a bulk data analysis. Your goal is to find the BEST search strategy through exploration and experimentation.
 
-"${question}"
+QUESTION TO ANSWER: "${question}"
+SEARCH SCOPE: ${path}
+
+YOUR TASK: Explore, experiment, then plan.
+
+STEP 1: EXPLORE THE REPOSITORY STRUCTURE
+Use listFiles to understand:
+- What types of files exist? (code, markdown, configs, etc.)
+- What are the naming patterns? (e.g., "Playbook", "Debrief", "spec", "test")
+- What directories and subdirectories exist?
+
+STEP 2: TEST DIFFERENT SEARCH QUERIES
+Run several experimental searches to find what works:
+- Try different keyword combinations related to the question
+- Test queries based on file naming patterns you discovered
+- Check which queries return relevant results vs empty results
+- Iterate until you find queries that actually return data
 
-Search scope: ${path}
+For example, if looking for customer data:
+- First try: search for "customer" - see what comes back
+- If empty, try: search for "playbook" or "debrief" (common doc names)
+- Check the actual content to understand terminology used
 
-Use attempt_completion to output your plan in this EXACT format:
+STEP 3: CREATE THE FINAL PLAN
+Based on your experiments, output the BEST search strategy.
 
-SEARCH_QUERY: <elasticsearch query using OR for multiple terms, quotes for exact phrases>
+Use attempt_completion with this EXACT format:
+
+SEARCH_QUERY: <the query that WORKED in your experiments - use OR for multiple terms>
 AGGREGATION: <summarize | list_unique | count | group_by>
 EXTRACTION_PROMPT: <what to extract from each search result>
-REASONING: <brief explanation>
-
-Example plan:
-SEARCH_QUERY: export OR function OR class OR tool
-AGGREGATION: list_unique
-EXTRACTION_PROMPT: Extract tool names and their purpose
-REASONING: Using list_unique to deduplicate tool definitions
+REASONING: <what you discovered, what queries you tested, why the final query works>
 
-IMPORTANT: Use attempt_completion immediately with your plan. Do NOT try to search or answer the question - just create the analysis plan.`;
+CRITICAL: Do NOT guess keywords. Actually run searches and see what returns results!`;
 
 	try {
-		// Planning phase - attempt_completion only
+		// Planning phase - full agent with exploration capabilities
 		const result = await delegate({
 			task: planningTask,
 			debug: options.debug,
@@ -383,9 +401,9 @@ IMPORTANT: Use attempt_completion immediately with your plan. Do NOT try to sear
 			tracer: options.tracer,
 			enableBash: false,
 			promptType: 'code-researcher',
-			allowedTools: [], // attempt_completion only (default tool)
-			maxIterations: 3,
-			timeout: 60
+			// Full tool access for exploration and experimentation
+			maxIterations: 15,
+			timeout: 180
 		});
 
 		const plan = parsePlanningResult(stripResultTags(result));

package/src/tools/common.js

@@ -47,7 +47,7 @@ export const bashSchema = z.object({
 });
 
 export const analyzeAllSchema = z.object({
-	question: z.string().min(1).describe('Free-form question to answer (e.g., "What features are customers using?", "List all API endpoints"). The AI will automatically plan the search strategy, process all matching data, and synthesize a comprehensive answer.'),
+	question: z.string().min(1).describe('Free-form question to answer (e.g., "What features are customers using?", "List all API endpoints"). The AI will automatically explore the repository, test search strategies, and synthesize a comprehensive answer.'),
 	path: z.string().optional().default('.').describe('Directory path to search in')
 });