@posthog/agent 1.16.6 → 1.17.0

package/dist/src/workflow/steps/research.js

@@ -3,11 +3,29 @@ import { RESEARCH_SYSTEM_PROMPT } from '../../agents/research.js';
 import { finalizeStepGitActions } from '../utils.js';
 
 const researchStep = async ({ step, context }) => {
-    const { task, cwd, isCloudMode, options, logger, fileManager, gitManager, promptBuilder, adapter, mcpServers, extractor, emitEvent, } = context;
+    const { task, cwd, isCloudMode, options, logger, fileManager, gitManager, promptBuilder, adapter, mcpServers, emitEvent, } = context;
     const stepLogger = logger.child('ResearchStep');
     const existingResearch = await fileManager.readResearch(task.id);
     if (existingResearch) {
-        stepLogger.info('Research already exists, skipping step', { taskId: task.id });
+        stepLogger.info('Research already exists', { taskId: task.id, hasQuestions: !!existingResearch.questions, answered: existingResearch.answered });
+        // If there are unanswered questions, re-emit them so UI can prompt user
+        if (existingResearch.questions && !existingResearch.answered) {
+            stepLogger.info('Re-emitting unanswered research questions', {
+                taskId: task.id,
+                questionCount: existingResearch.questions.length
+            });
+            emitEvent({
+                type: 'artifact',
+                ts: Date.now(),
+                kind: 'research_questions',
+                content: existingResearch.questions,
+            });
+            // In local mode, halt to allow user to answer
+            if (!isCloudMode) {
+                emitEvent(adapter.createStatusEvent('phase_complete', { phase: 'research' }));
+                return { status: 'skipped', halt: true };
+            }
+        }
         return { status: 'skipped' };
     }
     stepLogger.info('Starting research phase', { taskId: task.id });
@@ -37,7 +55,7 @@ const researchStep = async ({ step, context }) => {
         prompt: fullPrompt,
         options: { ...baseOptions, ...(options.queryOverrides || {}) },
     });
-    let researchContent = '';
+    let jsonContent = '';
     for await (const message of response) {
         emitEvent(adapter.createRawSDKEvent(message));
         const transformed = adapter.transform(message);
@@ -47,87 +65,103 @@ const researchStep = async ({ step, context }) => {
         if (message.type === 'assistant' && message.message?.content) {
             for (const c of message.message.content) {
                 if (c.type === 'text' && c.text) {
-                    researchContent += `${c.text}\n`;
+                    jsonContent += c.text;
                 }
             }
         }
     }
-    if (researchContent.trim()) {
-        await fileManager.writeResearch(task.id, researchContent.trim());
-        stepLogger.info('Research completed', { taskId: task.id });
+    if (!jsonContent.trim()) {
+        stepLogger.error('No JSON output from research agent', { taskId: task.id });
+        emitEvent({
+            type: 'error',
+            ts: Date.now(),
+            message: 'Research agent returned no output',
+        });
+        return { status: 'completed', halt: true };
+    }
+    // Parse JSON response
+    let evaluation;
+    try {
+        // Extract JSON from potential markdown code blocks or other wrapping
+        const jsonMatch = jsonContent.match(/\{[\s\S]*\}/);
+        if (!jsonMatch) {
+            throw new Error('No JSON object found in response');
+        }
+        evaluation = JSON.parse(jsonMatch[0]);
+        stepLogger.info('Parsed research evaluation', {
+            taskId: task.id,
+            score: evaluation.actionabilityScore,
+            hasQuestions: !!evaluation.questions,
+        });
+    }
+    catch (error) {
+        stepLogger.error('Failed to parse research JSON', {
+            taskId: task.id,
+            error: error instanceof Error ? error.message : String(error),
+            content: jsonContent.substring(0, 500),
+        });
+        emitEvent({
+            type: 'error',
+            ts: Date.now(),
+            message: `Failed to parse research JSON: ${error instanceof Error ? error.message : String(error)}`,
+        });
+        return { status: 'completed', halt: true };
+    }
+    // Add answered/answers fields to evaluation
+    if (evaluation.questions && evaluation.questions.length > 0) {
+        evaluation.answered = false;
+        evaluation.answers = undefined;
     }
+    // Always write research.json
+    await fileManager.writeResearch(task.id, evaluation);
+    stepLogger.info('Research evaluation written', {
+        taskId: task.id,
+        score: evaluation.actionabilityScore,
+        hasQuestions: !!evaluation.questions,
+    });
+    emitEvent({
+        type: 'artifact',
+        ts: Date.now(),
+        kind: 'research_evaluation',
+        content: evaluation,
+    });
     await gitManager.addAllPostHogFiles();
     await finalizeStepGitActions(context, step, {
         commitMessage: `Research phase for ${task.title}`,
     });
-    if (extractor && researchContent.trim()) {
-        try {
-            stepLogger.info('Extracting questions from research.md', { taskId: task.id });
-            const parsedQuestions = await extractor.extractQuestions(researchContent);
-            await fileManager.writeQuestions(task.id, {
-                questions: parsedQuestions,
-                answered: false,
-                answers: null,
-            });
-            emitEvent({
-                type: 'artifact',
-                ts: Date.now(),
-                kind: 'research_questions',
-                content: parsedQuestions,
-            });
-            stepLogger.info('Questions extracted successfully', {
-                taskId: task.id,
-                count: parsedQuestions.length,
-            });
-        }
-        catch (error) {
-            stepLogger.error('Failed to extract questions', {
-                taskId: task.id,
-                error: error instanceof Error ? error.message : String(error),
-            });
-            emitEvent({
-                type: 'error',
-                ts: Date.now(),
-                message: `Failed to extract questions: ${error instanceof Error ? error.message : String(error)}`,
-            });
-        }
-    }
-    else if (!extractor) {
-        stepLogger.warn('Question extractor not available, skipping question extraction. Ensure LLM gateway is configured.');
+    // Log whether questions need answering
+    if (evaluation.actionabilityScore < 0.7 && evaluation.questions && evaluation.questions.length > 0) {
+        stepLogger.info('Actionability score below threshold, questions needed', {
+            taskId: task.id,
+            score: evaluation.actionabilityScore,
+            questionCount: evaluation.questions.length,
+        });
         emitEvent({
-            type: 'status',
+            type: 'artifact',
             ts: Date.now(),
-            phase: 'extraction_skipped',
-            message: 'Question extraction skipped - extractor not configured',
+            kind: 'research_questions',
+            content: evaluation.questions,
         });
     }
+    else {
+        stepLogger.info('Actionability score acceptable, proceeding to planning', {
+            taskId: task.id,
+            score: evaluation.actionabilityScore,
+        });
+    }
+    // In local mode, always halt after research for user review
     if (!isCloudMode) {
         emitEvent(adapter.createStatusEvent('phase_complete', { phase: 'research' }));
         return { status: 'completed', halt: true };
     }
-    const questionsData = await fileManager.readQuestions(task.id);
-    if (questionsData && !questionsData.answered && extractor && researchContent.trim()) {
-        const researchQuestions = await extractor.extractQuestionsWithAnswers(researchContent);
-        const answers = researchQuestions.map((qa) => ({
-            questionId: qa.id,
-            selectedOption: qa.recommendedAnswer,
-            customInput: qa.justification,
-        }));
-        await fileManager.writeQuestions(task.id, {
-            questions: researchQuestions.map((qa) => ({
-                id: qa.id,
-                question: qa.question,
-                options: qa.options,
-            })),
-            answered: true,
-            answers,
-        });
-        await gitManager.addAllPostHogFiles();
-        await finalizeStepGitActions(context, step, {
-            commitMessage: `Answer research questions for ${task.title}`,
-        });
-        stepLogger.info('Auto-answered research questions', { taskId: task.id });
+    // In cloud mode, check if questions need answering
+    const researchData = await fileManager.readResearch(task.id);
+    if (researchData?.questions && !researchData.answered) {
+        // Questions need answering - halt for user input in cloud mode too
+        emitEvent(adapter.createStatusEvent('phase_complete', { phase: 'research' }));
+        return { status: 'completed', halt: true };
     }
+    // No questions or questions already answered - proceed to planning
     emitEvent(adapter.createStatusEvent('phase_complete', { phase: 'research' }));
     return { status: 'completed' };
 };

package/dist/src/workflow/steps/research.js.map

@@ -1 +1 @@
-{"version":3,"file":"research.js","sources":["../../../../src/workflow/steps/research.ts"],"sourcesContent":["import { query } from '@anthropic-ai/claude-agent-sdk';\nimport { RESEARCH_SYSTEM_PROMPT } from '../../agents/research.js';\nimport type { ExtractedQuestionWithAnswer } from '../../structured-extraction.js';\nimport type { WorkflowStepRunner } from '../types.js';\nimport { finalizeStepGitActions } from '../utils.js';\n\nexport const researchStep: WorkflowStepRunner = async ({ step, context }) => {\n    const {\n        task,\n        cwd,\n        isCloudMode,\n        options,\n        logger,\n        fileManager,\n        gitManager,\n        promptBuilder,\n        adapter,\n        mcpServers,\n        extractor,\n        emitEvent,\n    } = context;\n\n    const stepLogger = logger.child('ResearchStep');\n\n    const existingResearch = await fileManager.readResearch(task.id);\n    if (existingResearch) {\n        stepLogger.info('Research already exists, skipping step', { taskId: task.id });\n        return { status: 'skipped' };\n    }\n\n    stepLogger.info('Starting research phase', { taskId: task.id });\n    emitEvent(adapter.createStatusEvent('phase_start', { phase: 'research' }));\n\n    const researchPrompt = await promptBuilder.buildResearchPrompt(task, cwd);\n    const fullPrompt = `${RESEARCH_SYSTEM_PROMPT}\\n\\n${researchPrompt}`;\n\n    const baseOptions: Record<string, any> = {\n        model: step.model,\n        cwd,\n        permissionMode: 'plan',\n        settingSources: ['local'],\n        mcpServers,\n        // Allow research tools: read-only operations, web search, and MCP resources\n        allowedTools: [\n            'Read',\n            'Glob',\n            'Grep',\n            'WebFetch',\n            'WebSearch',\n            'ListMcpResources',\n            'ReadMcpResource',\n            'TodoWrite',\n            'BashOutput',\n        ],\n    };\n\n    const response = query({\n        prompt: fullPrompt,\n        options: { ...baseOptions, ...(options.queryOverrides || {}) },\n    });\n\n    let researchContent = '';\n    for await (const message of response) {\n        emitEvent(adapter.createRawSDKEvent(message));\n        const transformed = adapter.transform(message);\n        if (transformed) {\n            emitEvent(transformed);\n        }\n        if (message.type === 'assistant' && message.message?.content) {\n            for (const c of message.message.content) {\n                if (c.type === 'text' && c.text) {\n                    researchContent += `${c.text}\\n`;\n                }\n            }\n        }\n    }\n\n    if (researchContent.trim()) {\n        await fileManager.writeResearch(task.id, researchContent.trim());\n        stepLogger.info('Research completed', { taskId: task.id });\n    }\n\n    await gitManager.addAllPostHogFiles();\n    await finalizeStepGitActions(context, step, {\n        commitMessage: `Research phase for ${task.title}`,\n    });\n\n    if (extractor && researchContent.trim()) {\n        try {\n            stepLogger.info('Extracting questions from research.md', { taskId: task.id });\n            const parsedQuestions = await extractor.extractQuestions(researchContent);\n\n            await fileManager.writeQuestions(task.id, {\n                questions: parsedQuestions,\n                answered: false,\n                answers: null,\n            });\n\n            emitEvent({\n                type: 'artifact',\n                ts: Date.now(),\n                kind: 'research_questions',\n                content: parsedQuestions,\n            });\n\n            stepLogger.info('Questions extracted successfully', {\n                taskId: task.id,\n                count: parsedQuestions.length,\n            });\n        } catch (error) {\n            stepLogger.error('Failed to extract questions', {\n                taskId: task.id,\n                error: error instanceof Error ? error.message : String(error),\n            });\n            emitEvent({\n                type: 'error',\n                ts: Date.now(),\n                message: `Failed to extract questions: ${\n                    error instanceof Error ? error.message : String(error)\n                }`,\n            });\n        }\n    } else if (!extractor) {\n        stepLogger.warn(\n            'Question extractor not available, skipping question extraction. Ensure LLM gateway is configured.'\n        );\n        emitEvent({\n            type: 'status',\n            ts: Date.now(),\n            phase: 'extraction_skipped',\n            message: 'Question extraction skipped - extractor not configured',\n        });\n    }\n\n    if (!isCloudMode) {\n        emitEvent(adapter.createStatusEvent('phase_complete', { phase: 'research' }));\n        return { status: 'completed', halt: true };\n    }\n\n    const questionsData = await fileManager.readQuestions(task.id);\n    if (questionsData && !questionsData.answered && extractor && researchContent.trim()) {\n        const researchQuestions = await extractor.extractQuestionsWithAnswers(researchContent);\n        const answers = (researchQuestions as ExtractedQuestionWithAnswer[]).map((qa) => ({\n            questionId: qa.id,\n            selectedOption: qa.recommendedAnswer,\n            customInput: qa.justification,\n        }));\n\n        await fileManager.writeQuestions(task.id, {\n            questions: researchQuestions.map((qa) => ({\n                id: qa.id,\n                question: qa.question,\n                options: qa.options,\n            })),\n            answered: true,\n            answers,\n        });\n\n        await gitManager.addAllPostHogFiles();\n        await finalizeStepGitActions(context, step, {\n            commitMessage: `Answer research questions for ${task.title}`,\n        });\n        stepLogger.info('Auto-answered research questions', { taskId: task.id });\n    }\n\n    emitEvent(adapter.createStatusEvent('phase_complete', { phase: 'research' }));\n    return { status: 'completed' };\n};\n"],"names":[],"mappings":";;;;AAMO,MAAM,YAAY,GAAuB,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,KAAI;IACxE,MAAM,EACF,IAAI,EACJ,GAAG,EACH,WAAW,EACX,OAAO,EACP,MAAM,EACN,WAAW,EACX,UAAU,EACV,aAAa,EACb,OAAO,EACP,UAAU,EACV,SAAS,EACT,SAAS,GACZ,GAAG,OAAO;IAEX,MAAM,UAAU,GAAG,MAAM,CAAC,KAAK,CAAC,cAAc,CAAC;IAE/C,MAAM,gBAAgB,GAAG,MAAM,WAAW,CAAC,YAAY,CAAC,IAAI,CAAC,EAAE,CAAC;IAChE,IAAI,gBAAgB,EAAE;AAClB,QAAA,UAAU,CAAC,IAAI,CAAC,wCAAwC,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC;AAC9E,QAAA,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE;IAChC;AAEA,IAAA,UAAU,CAAC,IAAI,CAAC,yBAAyB,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC;AAC/D,IAAA,SAAS,CAAC,OAAO,CAAC,iBAAiB,CAAC,aAAa,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;IAE1E,MAAM,cAAc,GAAG,MAAM,aAAa,CAAC,mBAAmB,CAAC,IAAI,EAAE,GAAG,CAAC;AACzE,IAAA,MAAM,UAAU,GAAG,CAAA,EAAG,sBAAsB,CAAA,IAAA,EAAO,cAAc,EAAE;AAEnE,IAAA,MAAM,WAAW,GAAwB;QACrC,KAAK,EAAE,IAAI,CAAC,KAAK;QACjB,GAAG;AACH,QAAA,cAAc,EAAE,MAAM;QACtB,cAAc,EAAE,CAAC,OAAO,CAAC;QACzB,UAAU;;AAEV,QAAA,YAAY,EAAE;YACV,MAAM;YACN,MAAM;YACN,MAAM;YACN,UAAU;YACV,WAAW;YACX,kBAAkB;YAClB,iBAAiB;YACjB,WAAW;YACX,YAAY;AACf,SAAA;KACJ;IAED,MAAM,QAAQ,GAAG,KAAK,CAAC;AACnB,QAAA,MAAM,EAAE,UAAU;AAClB,QAAA,OAAO,EAAE,EAAE,GAAG,WAAW,EAAE,IAAI,OAAO,CAAC,cAAc,IAAI,EAAE,CAAC,EAAE;AACjE,KAAA,CAAC;IAEF,IAAI,eAAe,GAAG,EAAE;AACxB,IAAA,WAAW,MAAM,OAAO,IAAI,QAAQ,EAAE;QAClC,SAAS,CAAC,OAAO,CAAC,iBAAiB,CAAC,OAAO,CAAC,CAAC;QAC7C,MAAM,WAAW,GAAG,OAAO,CAAC,SAAS,CAAC,OAAO,CAAC;QAC9C,IAAI,WAAW,EAAE;YACb,SAAS,CAAC,WAAW,CAAC;QAC1B;AACA,QAAA,IAAI,OAAO,CAAC,IAAI,KAAK,WAAW,IAAI,OAAO,CAAC,OAAO,EAAE,OAAO,EAAE;YAC1D,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,OAAO,CAAC,OAAO,EAAE;gBACrC,IAAI,CAAC,CAAC,IAAI,KAAK,MAAM,IAAI,CAAC,CAAC,IAAI,EAAE;AAC7B,oBAAA,eAAe,IAAI,CAAA,EAAG,CAAC,CAAC,IAAI,IAAI;gBACpC;YACJ;QACJ;IACJ;AAEA,IAAA,IAAI,eAAe,CAAC,IAAI,EAAE,EAAE;AACxB,QAAA,MAAM,WAAW,CAAC,aAAa,CAAC,IAAI,CAAC,EAAE,EAAE,eAAe,CAAC,IAAI,EAAE,CAAC;AAChE,QAAA,UAAU,CAAC,IAAI,CAAC,oBAAoB,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC;IAC9D;AAEA,IAAA,MAAM,UAAU,CAAC,kBAAkB,EAAE;AACrC,IAAA,MAAM,sBAAsB,CAAC,OAAO,EAAE,IAAI,EAAE;AACxC,QAAA,aAAa,EAAE,CAAA,mBAAA,EAAsB,IAAI,CAAC,KAAK,CAAA,CAAE;AACpD,KAAA,CAAC;AAEF,IAAA,IAAI,SAAS,IAAI,eAAe,CAAC,IAAI,EAAE,EAAE;AACrC,QAAA,IAAI;AACA,YAAA,UAAU,CAAC,IAAI,CAAC,uCAAuC,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC;YAC7E,MAAM,eAAe,GAAG,MAAM,SAAS,CAAC,gBAAgB,CAAC,eAAe,CAAC;AAEzE,YAAA,MAAM,WAAW,CAAC,cAAc,CAAC,IAAI,CAAC,EAAE,EAAE;AACtC,gBAAA,SAAS,EAAE,eAAe;AAC1B,gBAAA,QAAQ,EAAE,KAAK;AACf,gBAAA,OAAO,EAAE,IAAI;AAChB,aAAA,CAAC;AAEF,YAAA,SAAS,CAAC;AACN,gBAAA,IAAI,EAAE,UAAU;AAChB,gBAAA,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE;AACd,gBAAA,IAAI,EAAE,oBAAoB;AAC1B,gBAAA,OAAO,EAAE,eAAe;AAC3B,aAAA,CAAC;AAEF,YAAA,UAAU,CAAC,IAAI,CAAC,kCAAkC,EAAE;gBAChD,MAAM,EAAE,IAAI,CAAC,EAAE;gBACf,KAAK,EAAE,eAAe,CAAC,MAAM;AAChC,aAAA,CAAC;QACN;QAAE,OAAO,KAAK,EAAE;AACZ,YAAA,UAAU,CAAC,KAAK,CAAC,6BAA6B,EAAE;gBAC5C,MAAM,EAAE,IAAI,CAAC,EAAE;AACf,gBAAA,KAAK,EAAE,KAAK,YAAY,KAAK,GAAG,KAAK,CAAC,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC;AAChE,aAAA,CAAC;AACF,YAAA,SAAS,CAAC;AACN,gBAAA,IAAI,EAAE,OAAO;AACb,gBAAA,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE;AACd,gBAAA,OAAO,EAAE,CAAA,6BAAA,EACL,KAAK,YAAY,KAAK,GAAG,KAAK,CAAC,OAAO,GAAG,MAAM,CAAC,KAAK,CACzD,CAAA,CAAE;AACL,aAAA,CAAC;QACN;IACJ;SAAO,IAAI,CAAC,SAAS,EAAE;AACnB,QAAA,UAAU,CAAC,IAAI,CACX,mGAAmG,CACtG;AACD,QAAA,SAAS,CAAC;AACN,YAAA,IAAI,EAAE,QAAQ;AACd,YAAA,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE;AACd,YAAA,KAAK,EAAE,oBAAoB;AAC3B,YAAA,OAAO,EAAE,wDAAwD;AACpE,SAAA,CAAC;IACN;IAEA,IAAI,CAAC,WAAW,EAAE;AACd,QAAA,SAAS,CAAC,OAAO,CAAC,iBAAiB,CAAC,gBAAgB,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;QAC7E,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE;IAC9C;IAEA,MAAM,aAAa,GAAG,MAAM,WAAW,CAAC,aAAa,CAAC,IAAI,CAAC,EAAE,CAAC;AAC9D,IAAA,IAAI,aAAa,IAAI,CAAC,aAAa,CAAC,QAAQ,IAAI,SAAS,IAAI,eAAe,CAAC,IAAI,EAAE,EAAE;QACjF,MAAM,iBAAiB,GAAG,MAAM,SAAS,CAAC,2BAA2B,CAAC,eAAe,CAAC;QACtF,MAAM,OAAO,GAAI,iBAAmD,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM;YAC9E,UAAU,EAAE,EAAE,CAAC,EAAE;YACjB,cAAc,EAAE,EAAE,CAAC,iBAAiB;YACpC,WAAW,EAAE,EAAE,CAAC,aAAa;AAChC,SAAA,CAAC,CAAC;AAEH,QAAA,MAAM,WAAW,CAAC,cAAc,CAAC,IAAI,CAAC,EAAE,EAAE;YACtC,SAAS,EAAE,iBAAiB,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM;gBACtC,EAAE,EAAE,EAAE,CAAC,EAAE;gBACT,QAAQ,EAAE,EAAE,CAAC,QAAQ;gBACrB,OAAO,EAAE,EAAE,CAAC,OAAO;AACtB,aAAA,CAAC,CAAC;AACH,YAAA,QAAQ,EAAE,IAAI;YACd,OAAO;AACV,SAAA,CAAC;AAEF,QAAA,MAAM,UAAU,CAAC,kBAAkB,EAAE;AACrC,QAAA,MAAM,sBAAsB,CAAC,OAAO,EAAE,IAAI,EAAE;AACxC,YAAA,aAAa,EAAE,CAAA,8BAAA,EAAiC,IAAI,CAAC,KAAK,CAAA,CAAE;AAC/D,SAAA,CAAC;AACF,QAAA,UAAU,CAAC,IAAI,CAAC,kCAAkC,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC;IAC5E;AAEA,IAAA,SAAS,CAAC,OAAO,CAAC,iBAAiB,CAAC,gBAAgB,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;AAC7E,IAAA,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE;AAClC;;;;"}
+{"version":3,"file":"research.js","sources":["../../../../src/workflow/steps/research.ts"],"sourcesContent":["import { query } from '@anthropic-ai/claude-agent-sdk';\nimport { RESEARCH_SYSTEM_PROMPT } from '../../agents/research.js';\nimport type { WorkflowStepRunner } from '../types.js';\nimport type { ResearchEvaluation } from '../../types.js';\nimport { finalizeStepGitActions } from '../utils.js';\n\nexport const researchStep: WorkflowStepRunner = async ({ step, context }) => {\n    const {\n        task,\n        cwd,\n        isCloudMode,\n        options,\n        logger,\n        fileManager,\n        gitManager,\n        promptBuilder,\n        adapter,\n        mcpServers,\n        emitEvent,\n    } = context;\n\n    const stepLogger = logger.child('ResearchStep');\n\n    const existingResearch = await fileManager.readResearch(task.id);\n    if (existingResearch) {\n        stepLogger.info('Research already exists', { taskId: task.id, hasQuestions: !!existingResearch.questions, answered: existingResearch.answered });\n        \n        // If there are unanswered questions, re-emit them so UI can prompt user\n        if (existingResearch.questions && !existingResearch.answered) {\n            stepLogger.info('Re-emitting unanswered research questions', { \n                taskId: task.id,\n                questionCount: existingResearch.questions.length \n            });\n            \n            emitEvent({\n                type: 'artifact',\n                ts: Date.now(),\n                kind: 'research_questions',\n                content: existingResearch.questions,\n            });\n            \n            // In local mode, halt to allow user to answer\n            if (!isCloudMode) {\n                emitEvent(adapter.createStatusEvent('phase_complete', { phase: 'research' }));\n                return { status: 'skipped', halt: true };\n            }\n        }\n        \n        return { status: 'skipped' };\n    }\n\n    stepLogger.info('Starting research phase', { taskId: task.id });\n    emitEvent(adapter.createStatusEvent('phase_start', { phase: 'research' }));\n\n    const researchPrompt = await promptBuilder.buildResearchPrompt(task, cwd);\n    const fullPrompt = `${RESEARCH_SYSTEM_PROMPT}\\n\\n${researchPrompt}`;\n\n    const baseOptions: Record<string, any> = {\n        model: step.model,\n        cwd,\n        permissionMode: 'plan',\n        settingSources: ['local'],\n        mcpServers,\n        // Allow research tools: read-only operations, web search, and MCP resources\n        allowedTools: [\n            'Read',\n            'Glob',\n            'Grep',\n            'WebFetch',\n            'WebSearch',\n            'ListMcpResources',\n            'ReadMcpResource',\n            'TodoWrite',\n            'BashOutput',\n        ],\n    };\n\n    const response = query({\n        prompt: fullPrompt,\n        options: { ...baseOptions, ...(options.queryOverrides || {}) },\n    });\n\n    let jsonContent = '';\n    for await (const message of response) {\n        emitEvent(adapter.createRawSDKEvent(message));\n        const transformed = adapter.transform(message);\n        if (transformed) {\n            emitEvent(transformed);\n        }\n        if (message.type === 'assistant' && message.message?.content) {\n            for (const c of message.message.content) {\n                if (c.type === 'text' && c.text) {\n                    jsonContent += c.text;\n                }\n            }\n        }\n    }\n\n    if (!jsonContent.trim()) {\n        stepLogger.error('No JSON output from research agent', { taskId: task.id });\n        emitEvent({\n            type: 'error',\n            ts: Date.now(),\n            message: 'Research agent returned no output',\n        });\n        return { status: 'completed', halt: true };\n    }\n\n    // Parse JSON response\n    let evaluation: ResearchEvaluation;\n    try {\n        // Extract JSON from potential markdown code blocks or other wrapping\n        const jsonMatch = jsonContent.match(/\\{[\\s\\S]*\\}/);\n        if (!jsonMatch) {\n            throw new Error('No JSON object found in response');\n        }\n        evaluation = JSON.parse(jsonMatch[0]);\n        stepLogger.info('Parsed research evaluation', {\n            taskId: task.id,\n            score: evaluation.actionabilityScore,\n            hasQuestions: !!evaluation.questions,\n        });\n    } catch (error) {\n        stepLogger.error('Failed to parse research JSON', {\n            taskId: task.id,\n            error: error instanceof Error ? error.message : String(error),\n            content: jsonContent.substring(0, 500),\n        });\n        emitEvent({\n            type: 'error',\n            ts: Date.now(),\n            message: `Failed to parse research JSON: ${\n                error instanceof Error ? error.message : String(error)\n            }`,\n        });\n        return { status: 'completed', halt: true };\n    }\n\n    // Add answered/answers fields to evaluation\n    if (evaluation.questions && evaluation.questions.length > 0) {\n        evaluation.answered = false;\n        evaluation.answers = undefined;\n    }\n\n    // Always write research.json\n    await fileManager.writeResearch(task.id, evaluation);\n    stepLogger.info('Research evaluation written', {\n        taskId: task.id,\n        score: evaluation.actionabilityScore,\n        hasQuestions: !!evaluation.questions,\n    });\n\n    emitEvent({\n        type: 'artifact',\n        ts: Date.now(),\n        kind: 'research_evaluation',\n        content: evaluation,\n    });\n\n    await gitManager.addAllPostHogFiles();\n    await finalizeStepGitActions(context, step, {\n        commitMessage: `Research phase for ${task.title}`,\n    });\n\n    // Log whether questions need answering\n    if (evaluation.actionabilityScore < 0.7 && evaluation.questions && evaluation.questions.length > 0) {\n        stepLogger.info('Actionability score below threshold, questions needed', {\n            taskId: task.id,\n            score: evaluation.actionabilityScore,\n            questionCount: evaluation.questions.length,\n        });\n        \n        emitEvent({\n            type: 'artifact',\n            ts: Date.now(),\n            kind: 'research_questions',\n            content: evaluation.questions,\n        });\n    } else {\n        stepLogger.info('Actionability score acceptable, proceeding to planning', {\n            taskId: task.id,\n            score: evaluation.actionabilityScore,\n        });\n    }\n\n    // In local mode, always halt after research for user review\n    if (!isCloudMode) {\n        emitEvent(adapter.createStatusEvent('phase_complete', { phase: 'research' }));\n        return { status: 'completed', halt: true };\n    }\n\n    // In cloud mode, check if questions need answering\n    const researchData = await fileManager.readResearch(task.id);\n    if (researchData?.questions && !researchData.answered) {\n        // Questions need answering - halt for user input in cloud mode too\n        emitEvent(adapter.createStatusEvent('phase_complete', { phase: 'research' }));\n        return { status: 'completed', halt: true };\n    }\n\n    // No questions or questions already answered - proceed to planning\n    emitEvent(adapter.createStatusEvent('phase_complete', { phase: 'research' }));\n    return { status: 'completed' };\n};\n"],"names":[],"mappings":";;;;AAMO,MAAM,YAAY,GAAuB,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,KAAI;IACxE,MAAM,EACF,IAAI,EACJ,GAAG,EACH,WAAW,EACX,OAAO,EACP,MAAM,EACN,WAAW,EACX,UAAU,EACV,aAAa,EACb,OAAO,EACP,UAAU,EACV,SAAS,GACZ,GAAG,OAAO;IAEX,MAAM,UAAU,GAAG,MAAM,CAAC,KAAK,CAAC,cAAc,CAAC;IAE/C,MAAM,gBAAgB,GAAG,MAAM,WAAW,CAAC,YAAY,CAAC,IAAI,CAAC,EAAE,CAAC;IAChE,IAAI,gBAAgB,EAAE;QAClB,UAAU,CAAC,IAAI,CAAC,yBAAyB,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,EAAE,YAAY,EAAE,CAAC,CAAC,gBAAgB,CAAC,SAAS,EAAE,QAAQ,EAAE,gBAAgB,CAAC,QAAQ,EAAE,CAAC;;QAGhJ,IAAI,gBAAgB,CAAC,SAAS,IAAI,CAAC,gBAAgB,CAAC,QAAQ,EAAE;AAC1D,YAAA,UAAU,CAAC,IAAI,CAAC,2CAA2C,EAAE;gBACzD,MAAM,EAAE,IAAI,CAAC,EAAE;AACf,gBAAA,aAAa,EAAE,gBAAgB,CAAC,SAAS,CAAC;AAC7C,aAAA,CAAC;AAEF,YAAA,SAAS,CAAC;AACN,gBAAA,IAAI,EAAE,UAAU;AAChB,gBAAA,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE;AACd,gBAAA,IAAI,EAAE,oBAAoB;gBAC1B,OAAO,EAAE,gBAAgB,CAAC,SAAS;AACtC,aAAA,CAAC;;YAGF,IAAI,CAAC,WAAW,EAAE;AACd,gBAAA,SAAS,CAAC,OAAO,CAAC,iBAAiB,CAAC,gBAAgB,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;gBAC7E,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE;YAC5C;QACJ;AAEA,QAAA,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE;IAChC;AAEA,IAAA,UAAU,CAAC,IAAI,CAAC,yBAAyB,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC;AAC/D,IAAA,SAAS,CAAC,OAAO,CAAC,iBAAiB,CAAC,aAAa,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;IAE1E,MAAM,cAAc,GAAG,MAAM,aAAa,CAAC,mBAAmB,CAAC,IAAI,EAAE,GAAG,CAAC;AACzE,IAAA,MAAM,UAAU,GAAG,CAAA,EAAG,sBAAsB,CAAA,IAAA,EAAO,cAAc,EAAE;AAEnE,IAAA,MAAM,WAAW,GAAwB;QACrC,KAAK,EAAE,IAAI,CAAC,KAAK;QACjB,GAAG;AACH,QAAA,cAAc,EAAE,MAAM;QACtB,cAAc,EAAE,CAAC,OAAO,CAAC;QACzB,UAAU;;AAEV,QAAA,YAAY,EAAE;YACV,MAAM;YACN,MAAM;YACN,MAAM;YACN,UAAU;YACV,WAAW;YACX,kBAAkB;YAClB,iBAAiB;YACjB,WAAW;YACX,YAAY;AACf,SAAA;KACJ;IAED,MAAM,QAAQ,GAAG,KAAK,CAAC;AACnB,QAAA,MAAM,EAAE,UAAU;AAClB,QAAA,OAAO,EAAE,EAAE,GAAG,WAAW,EAAE,IAAI,OAAO,CAAC,cAAc,IAAI,EAAE,CAAC,EAAE;AACjE,KAAA,CAAC;IAEF,IAAI,WAAW,GAAG,EAAE;AACpB,IAAA,WAAW,MAAM,OAAO,IAAI,QAAQ,EAAE;QAClC,SAAS,CAAC,OAAO,CAAC,iBAAiB,CAAC,OAAO,CAAC,CAAC;QAC7C,MAAM,WAAW,GAAG,OAAO,CAAC,SAAS,CAAC,OAAO,CAAC;QAC9C,IAAI,WAAW,EAAE;YACb,SAAS,CAAC,WAAW,CAAC;QAC1B;AACA,QAAA,IAAI,OAAO,CAAC,IAAI,KAAK,WAAW,IAAI,OAAO,CAAC,OAAO,EAAE,OAAO,EAAE;YAC1D,KAAK,MAAM,CAAC,IAAI,OAAO,CAAC,OAAO,CAAC,OAAO,EAAE;gBACrC,IAAI,CAAC,CAAC,IAAI,KAAK,MAAM,IAAI,CAAC,CAAC,IAAI,EAAE;AAC7B,oBAAA,WAAW,IAAI,CAAC,CAAC,IAAI;gBACzB;YACJ;QACJ;IACJ;AAEA,IAAA,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,EAAE;AACrB,QAAA,UAAU,CAAC,KAAK,CAAC,oCAAoC,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC;AAC3E,QAAA,SAAS,CAAC;AACN,YAAA,IAAI,EAAE,OAAO;AACb,YAAA,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE;AACd,YAAA,OAAO,EAAE,mCAAmC;AAC/C,SAAA,CAAC;QACF,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE;IAC9C;;AAGA,IAAA,IAAI,UAA8B;AAClC,IAAA,IAAI;;QAEA,MAAM,SAAS,GAAG,WAAW,CAAC,KAAK,CAAC,aAAa,CAAC;QAClD,IAAI,CAAC,SAAS,EAAE;AACZ,YAAA,MAAM,IAAI,KAAK,CAAC,kCAAkC,CAAC;QACvD;QACA,UAAU,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC;AACrC,QAAA,UAAU,CAAC,IAAI,CAAC,4BAA4B,EAAE;YAC1C,MAAM,EAAE,IAAI,CAAC,EAAE;YACf,KAAK,EAAE,UAAU,CAAC,kBAAkB;AACpC,YAAA,YAAY,EAAE,CAAC,CAAC,UAAU,CAAC,SAAS;AACvC,SAAA,CAAC;IACN;IAAE,OAAO,KAAK,EAAE;AACZ,QAAA,UAAU,CAAC,KAAK,CAAC,+BAA+B,EAAE;YAC9C,MAAM,EAAE,IAAI,CAAC,EAAE;AACf,YAAA,KAAK,EAAE,KAAK,YAAY,KAAK,GAAG,KAAK,CAAC,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC;YAC7D,OAAO,EAAE,WAAW,CAAC,SAAS,CAAC,CAAC,EAAE,GAAG,CAAC;AACzC,SAAA,CAAC;AACF,QAAA,SAAS,CAAC;AACN,YAAA,IAAI,EAAE,OAAO;AACb,YAAA,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE;AACd,YAAA,OAAO,EAAE,CAAA,+BAAA,EACL,KAAK,YAAY,KAAK,GAAG,KAAK,CAAC,OAAO,GAAG,MAAM,CAAC,KAAK,CACzD,CAAA,CAAE;AACL,SAAA,CAAC;QACF,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE;IAC9C;;AAGA,IAAA,IAAI,UAAU,CAAC,SAAS,IAAI,UAAU,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE;AACzD,QAAA,UAAU,CAAC,QAAQ,GAAG,KAAK;AAC3B,QAAA,UAAU,CAAC,OAAO,GAAG,SAAS;IAClC;;IAGA,MAAM,WAAW,CAAC,aAAa,CAAC,IAAI,CAAC,EAAE,EAAE,UAAU,CAAC;AACpD,IAAA,UAAU,CAAC,IAAI,CAAC,6BAA6B,EAAE;QAC3C,MAAM,EAAE,IAAI,CAAC,EAAE;QACf,KAAK,EAAE,UAAU,CAAC,kBAAkB;AACpC,QAAA,YAAY,EAAE,CAAC,CAAC,UAAU,CAAC,SAAS;AACvC,KAAA,CAAC;AAEF,IAAA,SAAS,CAAC;AACN,QAAA,IAAI,EAAE,UAAU;AAChB,QAAA,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE;AACd,QAAA,IAAI,EAAE,qBAAqB;AAC3B,QAAA,OAAO,EAAE,UAAU;AACtB,KAAA,CAAC;AAEF,IAAA,MAAM,UAAU,CAAC,kBAAkB,EAAE;AACrC,IAAA,MAAM,sBAAsB,CAAC,OAAO,EAAE,IAAI,EAAE;AACxC,QAAA,aAAa,EAAE,CAAA,mBAAA,EAAsB,IAAI,CAAC,KAAK,CAAA,CAAE;AACpD,KAAA,CAAC;;AAGF,IAAA,IAAI,UAAU,CAAC,kBAAkB,GAAG,GAAG,IAAI,UAAU,CAAC,SAAS,IAAI,UAAU,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE;AAChG,QAAA,UAAU,CAAC,IAAI,CAAC,uDAAuD,EAAE;YACrE,MAAM,EAAE,IAAI,CAAC,EAAE;YACf,KAAK,EAAE,UAAU,CAAC,kBAAkB;AACpC,YAAA,aAAa,EAAE,UAAU,CAAC,SAAS,CAAC,MAAM;AAC7C,SAAA,CAAC;AAEF,QAAA,SAAS,CAAC;AACN,YAAA,IAAI,EAAE,UAAU;AAChB,YAAA,EAAE,EAAE,IAAI,CAAC,GAAG,EAAE;AACd,YAAA,IAAI,EAAE,oBAAoB;YAC1B,OAAO,EAAE,UAAU,CAAC,SAAS;AAChC,SAAA,CAAC;IACN;SAAO;AACH,QAAA,UAAU,CAAC,IAAI,CAAC,wDAAwD,EAAE;YACtE,MAAM,EAAE,IAAI,CAAC,EAAE;YACf,KAAK,EAAE,UAAU,CAAC,kBAAkB;AACvC,SAAA,CAAC;IACN;;IAGA,IAAI,CAAC,WAAW,EAAE;AACd,QAAA,SAAS,CAAC,OAAO,CAAC,iBAAiB,CAAC,gBAAgB,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;QAC7E,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE;IAC9C;;IAGA,MAAM,YAAY,GAAG,MAAM,WAAW,CAAC,YAAY,CAAC,IAAI,CAAC,EAAE,CAAC;IAC5D,IAAI,YAAY,EAAE,SAAS,IAAI,CAAC,YAAY,CAAC,QAAQ,EAAE;;AAEnD,QAAA,SAAS,CAAC,OAAO,CAAC,iBAAiB,CAAC,gBAAgB,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;QAC7E,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,EAAE;IAC9C;;AAGA,IAAA,SAAS,CAAC,OAAO,CAAC,iBAAiB,CAAC,gBAAgB,EAAE,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;AAC7E,IAAA,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE;AAClC;;;;"}

package/dist/src/workflow/types.d.ts

@@ -6,7 +6,6 @@ import type { PromptBuilder } from '../prompt-builder.js';
 import type { TaskProgressReporter } from '../task-progress-reporter.js';
 import type { ProviderAdapter } from '../adapters/types.js';
 import type { PostHogAPIClient } from '../posthog-api.js';
-import type { StructuredExtractor } from '../structured-extraction.js';
 export interface WorkflowRuntime {
     task: Task;
     taskSlug: string;
@@ -21,7 +20,6 @@ export interface WorkflowRuntime {
     adapter: ProviderAdapter;
     mcpServers?: Record<string, any>;
     posthogAPI?: PostHogAPIClient;
-    extractor?: StructuredExtractor;
     emitEvent: (event: any) => void;
     stepResults: Record<string, any>;
 }

package/dist/src/workflow/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/workflow/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,oBAAoB,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAC9E,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,oBAAoB,CAAC;AACjD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,oBAAoB,CAAC;AAC7D,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AAC1D,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,8BAA8B,CAAC;AACzE,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAC;AAC5D,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAC1D,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,6BAA6B,CAAC;AAEvE,MAAM,WAAW,eAAe;IAC5B,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,GAAG,EAAE,MAAM,CAAC;IACZ,WAAW,EAAE,OAAO,CAAC;IACrB,OAAO,EAAE,oBAAoB,CAAC;IAC9B,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,kBAAkB,CAAC;IAChC,UAAU,EAAE,UAAU,CAAC;IACvB,aAAa,EAAE,aAAa,CAAC;IAC7B,gBAAgB,EAAE,oBAAoB,CAAC;IACvC,OAAO,EAAE,eAAe,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IACjC,UAAU,CAAC,EAAE,gBAAgB,CAAC;IAC9B,SAAS,CAAC,EAAE,mBAAmB,CAAC;IAChC,SAAS,EAAE,CAAC,KAAK,EAAE,GAAG,KAAK,IAAI,CAAC;IAChC,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;CACpC;AAED,MAAM,WAAW,sBAAsB;IACnC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,CAAC,EAAE,cAAc,GAAG,MAAM,CAAC;IACzC,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,GAAG,EAAE,kBAAkB,CAAC;CAC3B;AAED,MAAM,WAAW,mBAAmB;IAChC,IAAI,EAAE,sBAAsB,CAAC;IAC7B,OAAO,EAAE,eAAe,CAAC;CAC5B;AAED,MAAM,WAAW,kBAAkB;IAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,CAAC;IAChC,IAAI,CAAC,EAAE,OAAO,CAAC;CAClB;AAED,MAAM,MAAM,kBAAkB,GAAG,CAAC,OAAO,EAAE,mBAAmB,KAAK,OAAO,CAAC,kBAAkB,CAAC,CAAC;AAE/F,MAAM,MAAM,kBAAkB,GAAG,sBAAsB,EAAE,CAAC"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/workflow/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,IAAI,EAAE,oBAAoB,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAC9E,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,oBAAoB,CAAC;AACjD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,oBAAoB,CAAC;AAC7D,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AAC1D,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,8BAA8B,CAAC;AACzE,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,sBAAsB,CAAC;AAC5D,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAE1D,MAAM,WAAW,eAAe;IAC5B,IAAI,EAAE,IAAI,CAAC;IACX,QAAQ,EAAE,MAAM,CAAC;IACjB,GAAG,EAAE,MAAM,CAAC;IACZ,WAAW,EAAE,OAAO,CAAC;IACrB,OAAO,EAAE,oBAAoB,CAAC;IAC9B,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,kBAAkB,CAAC;IAChC,UAAU,EAAE,UAAU,CAAC;IACvB,aAAa,EAAE,aAAa,CAAC;IAC7B,gBAAgB,EAAE,oBAAoB,CAAC;IACvC,OAAO,EAAE,eAAe,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IACjC,UAAU,CAAC,EAAE,gBAAgB,CAAC;IAC9B,SAAS,EAAE,CAAC,KAAK,EAAE,GAAG,KAAK,IAAI,CAAC;IAChC,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;CACpC;AAED,MAAM,WAAW,sBAAsB;IACnC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,cAAc,CAAC,EAAE,cAAc,GAAG,MAAM,CAAC;IACzC,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,IAAI,CAAC,EAAE,OAAO,CAAC;IACf,GAAG,EAAE,kBAAkB,CAAC;CAC3B;AAED,MAAM,WAAW,mBAAmB;IAChC,IAAI,EAAE,sBAAsB,CAAC;IAC7B,OAAO,EAAE,eAAe,CAAC;CAC5B;AAED,MAAM,WAAW,kBAAkB;IAC/B,MAAM,EAAE,WAAW,GAAG,SAAS,CAAC;IAChC,IAAI,CAAC,EAAE,OAAO,CAAC;CAClB;AAED,MAAM,MAAM,kBAAkB,GAAG,CAAC,OAAO,EAAE,mBAAmB,KAAK,OAAO,CAAC,kBAAkB,CAAC,CAAC;AAE/F,MAAM,MAAM,kBAAkB,GAAG,sBAAsB,EAAE,CAAC"}

package/dist/templates/plan-template.md

@@ -38,8 +38,4 @@ path/to/existing/file.ts - Changes needed
 
 - Key architectural decisions
 - Potential risks and mitigation
-- Testing approach
-
----
-
-*Generated by PostHog Agent*
+- Testing approach

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@posthog/agent",
-  "version": "1.16.6",
+  "version": "1.17.0",
   "description": "TypeScript agent framework wrapping Claude Agent SDK with Git-based task execution for PostHog",
   "main": "./dist/index.js",
   "module": "./dist/index.js",
@@ -49,12 +49,8 @@
     "typescript": "^5.5.0"
   },
   "dependencies": {
-    "@ai-sdk/anthropic": "^1.0.6",
-    "@ai-sdk/openai": "^1.0.10",
     "@anthropic-ai/claude-agent-sdk": "^0.1.1",
-    "ai": "^4.2.0",
-    "dotenv": "^17.2.3",
-    "zod": "^3.24.1"
+    "dotenv": "^17.2.3"
   },
   "files": [
     "dist/**/*",

package/src/agent.ts

@@ -10,7 +10,6 @@ import type { ProviderAdapter } from './adapters/types.js';
 import { Logger } from './utils/logger.js';
 import { PromptBuilder } from './prompt-builder.js';
 import { TaskProgressReporter } from './task-progress-reporter.js';
-import { AISDKExtractor, type StructuredExtractor, type ExtractedQuestion, type ExtractedQuestionWithAnswer } from './structured-extraction.js';
 import { TASK_WORKFLOW } from './workflow/config.js';
 import type { WorkflowRuntime } from './workflow/types.js';
 
@@ -26,7 +25,6 @@ export class Agent {
     private logger: Logger;
     private progressReporter: TaskProgressReporter;
     private promptBuilder: PromptBuilder;
-    private extractor?: StructuredExtractor;
     private mcpServers?: Record<string, any>;
     private canUseTool?: CanUseTool;
     public debug: boolean;
@@ -92,7 +90,6 @@ export class Agent {
             logger: this.logger.child('PromptBuilder')
         });
         this.progressReporter = new TaskProgressReporter(this.posthogAPI, this.logger);
-        this.extractor = new AISDKExtractor({apiKey: config.posthogApiKey, gatewayUrl: this.posthogAPI?.getLlmGatewayUrl() ?? '', logger: this.logger.child('AISDKExtractor')});
     }
 
     /**
@@ -159,7 +156,6 @@ export class Agent {
                 adapter: this.adapter,
                 mcpServers: this.mcpServers,
                 posthogAPI: this.posthogAPI,
-                extractor: this.extractor,
                 emitEvent: (event: any) => this.emitEvent(event),
                 stepResults: {},
             };
@@ -283,32 +279,11 @@ export class Agent {
         return await this.fileManager.readPlan(taskId);
     }
 
-    async extractQuestionsFromResearch(taskId: string, includeAnswers: boolean = false): Promise<ExtractedQuestion[] | ExtractedQuestionWithAnswer[]> {
-        this.logger.info('Extracting questions from research.md', { taskId, includeAnswers });
-        
-        if (!this.extractor) {
-            throw new Error('OpenAI extractor not initialized. Ensure the LLM gateway is configured.');
-        }
-
-        const researchContent = await this.fileManager.readResearch(taskId);
-        if (!researchContent) {
-            throw new Error('research.md not found for task ' + taskId);
-        }
-
-        if (includeAnswers) {
-            return await this.extractor.extractQuestionsWithAnswers(researchContent);
-        } else {
-            return await this.extractor.extractQuestions(researchContent);
-        }
-    }
-
     // Git operations for task execution
     async createPlanningBranch(taskId: string): Promise<string> {
         this.logger.info('Creating planning branch', { taskId });
         const branchName = await this.gitManager.createTaskPlanningBranch(taskId);
         this.logger.debug('Planning branch created', { taskId, branchName });
-        // Only create gitignore after we're on the new branch
-        await this.fileManager.ensureGitignore();
         return branchName;
     }
     
@@ -422,12 +397,17 @@ Generated by PostHog Agent`;
             const branchName = await this.gitManager.createTaskBranch(taskSlug);
             this.emitEvent(this.adapter.createStatusEvent('branch_created', { branch: branchName }));
 
-            await this.fileManager.ensureGitignore();
             await this.gitManager.addAllPostHogFiles();
+            
+            // Only commit if there are changes or we're in cloud mode
             if (isCloudMode) {
                 await this.gitManager.commitAndPush(`Initialize task ${taskSlug}`, { allowEmpty: true });
             } else {
-                await this.gitManager.commitChanges(`Initialize task ${taskSlug}`);
+                // Check if there are any changes before committing
+                const hasChanges = await this.gitManager.hasStagedChanges();
+                if (hasChanges) {
+                    await this.gitManager.commitChanges(`Initialize task ${taskSlug}`);
+                }
             }
         } else {
             this.logger.info('Switching to existing task branch', { branch: existingBranch });
@@ -446,10 +426,6 @@ Generated by PostHog Agent`;
         if (resolvedToken) {
             process.env.OPENAI_API_KEY = resolvedToken;
         }
-
-        if (!this.extractor) {
-            this.extractor = new AISDKExtractor({apiKey: resolvedToken || '', gatewayUrl: resolvedGatewayUrl || '', logger: this.logger.child('AISDKExtractor')});
-        }
     }
 
     private async ensurePullRequest(task: Task, stepResults: Record<string, any>): Promise<void> {

package/src/agents/research.ts

@@ -1,80 +1,111 @@
 export const RESEARCH_SYSTEM_PROMPT = `<role>
-PostHog AI Research Agent — analyze codebases to understand implementation context and identify areas of focus for development tasks.
+PostHog AI Research Agent — analyze codebases to evaluate task actionability and identify missing information.
 </role>
 
 <constraints>
 - Read-only: analyze files, search code, explore structure
 - No modifications or code changes
+- Output structured JSON only
 </constraints>
 
 <objective>
-Your PRIMARY goal is to understand the codebase thoroughly and provide context for the planning phase.
+Your PRIMARY goal is to evaluate whether a task is actionable and assign an actionability score.
 
-ONLY generate clarifying questions if:
-- The task description is genuinely vague or ambiguous
-- There are multiple valid architectural approaches with significant tradeoffs
-- Critical information is missing that cannot be inferred from the codebase
+Calculate an actionabilityScore (0-1) based on:
+- **Task clarity** (0.4 weight): Is the task description specific and unambiguous?
+- **Codebase context** (0.3 weight): Can you locate the relevant code and patterns?
+- **Architectural decisions** (0.2 weight): Are the implementation approaches clear?
+- **Dependencies** (0.1 weight): Are required dependencies and constraints understood?
 
-DO NOT ask questions like "how should I fix this" or "what approach do you prefer" — that defeats the purpose of autonomous task execution. The user has already specified what they want done.
+If actionabilityScore < 0.7, generate specific clarifying questions to increase confidence.
+
+DO NOT ask questions like "how should I fix this" — focus on missing information that prevents confident planning.
 </objective>
 
 <process>
 1. Explore repository structure and identify relevant files/components
 2. Understand existing patterns, conventions, and dependencies
-3. Locate similar implementations or related code
-4. Identify the key areas of the codebase that will be affected
-5. Document your findings to provide context for planning
-6. ONLY if genuinely needed: generate 2-3 specific clarification questions
+3. Calculate actionabilityScore based on clarity, context, architecture, and dependencies
+4. Identify key files that will need modification
+5. If score < 0.7: generate 2-4 specific questions to resolve blockers
+6. Output JSON matching ResearchEvaluation schema
 </process>
 
 <output_format>
-Output ONLY the markdown artifact with no preamble:
-
-\`\`\`markdown
-# Research Findings
-
-## Codebase Analysis
-[Brief summary of relevant code structure, patterns, and files]
-
-## Key Areas of Focus
-[List specific files/components that need modification]
-
-## Implementation Context
-[Important patterns, dependencies, or constraints found in the code]
-
-## Clarifying Questions
-[ONLY include this section if it will increase the quality of the plan]
-
-## Question 1: [Specific architectural decision]
-**Options:**
-- a) [Concrete option with file references]
-- b) [Alternative with file references]
-- c) Something else (please specify)
-\`\`\`
-
-Format requirements:
-- Use "## Question N:" for question headers (h2)
-- Follow with "**Options:**" on its own line
-- Start options with "- a)", "- b)", "- c)"
-- Always include "c) Something else (please specify)"
-- Max 4 questions total
+Output ONLY valid JSON with no markdown wrappers, no preamble, no explanation:
+
+{
+  "actionabilityScore": 0.85,
+  "context": "Brief 2-3 sentence summary of the task and implementation approach",
+  "keyFiles": ["path/to/file1.ts", "path/to/file2.ts"],
+  "blockers": ["Optional: what's preventing full confidence"],
+  "questions": [
+    {
+      "id": "q1",
+      "question": "Specific architectural decision needed?",
+      "options": [
+        "First approach with concrete details",
+        "Alternative approach with concrete details",
+        "Third option if needed"
+      ]
+    }
+  ]
+}
+
+Rules:
+- actionabilityScore: number between 0 and 1
+- context: concise summary for planning phase
+- keyFiles: array of file paths that need modification
+- blockers: optional array explaining confidence gaps
+- questions: ONLY include if actionabilityScore < 0.7
+- Each question must have 2-3 options (maximum 3)
+- Max 3 questions total
 </output_format>
 
-<examples>
+<scoring_examples>
+<example score="0.9">
+Task: "Fix typo in login button text"
+Reasoning: Completely clear task, found exact component, no architectural decisions
+</example>
+
+<example score="0.75">
+Task: "Add caching to API endpoints"
+Reasoning: Clear goal, found endpoints, but multiple caching strategies possible
+</example>
+
+<example score="0.55">
+Task: "Improve performance"
+Reasoning: Vague task, unclear scope, needs questions about which areas to optimize
+Questions needed: Which features are slow? What metrics define success?
+</example>
+
+<example score="0.3">
+Task: "Add the new feature"
+Reasoning: Extremely vague, no context, cannot locate relevant code
+Questions needed: What feature? Which product area? What should it do?
+</example>
+</scoring_examples>
+
+<question_examples>
 <good_example>
-Task: "Fix authentication bug in login flow"
-Output: Research findings showing auth flow files, patterns used, NO questions needed
+{
+  "id": "q1",
+  "question": "Which caching layer should we use for API responses?",
+  "options": [
+    "Redis (existing infrastructure, requires setup)",
+    "In-memory cache (simpler, but not distributed)",
+    "Browser-side caching only (minimal backend changes)"
+  ]
+}
 </good_example>
 
 <bad_example>
-Task: "Fix authentication bug"
-Output: "How should I fix the authentication? a) Fix it one way b) Fix it another way"
-Reason: Don't ask HOW to do the task — that's what the agent is for
+{
+  "id": "q1", 
+  "question": "How should I implement this?",
+  "options": ["One way", "Another way"]
+}
+Reason: Too vague, doesn't explain the tradeoffs
 </bad_example>
-
-<good_example>
-Task: "Add caching to API endpoints"
-Output: Research showing existing cache implementations, question about cache backend choice IF multiple production systems are already in use
-</good_example>
-</examples>`;
+</question_examples>`;