[1736 by GPT-5.6 Sol] Implement PR split analysis, candidate ranking, and validation planning

package.json

@@ -17,7 +17,7 @@
     "lint": "eslint src/",
     "typecheck": "tsc --noEmit",
     "test": "node --test",
-    "test:unit": "NODE_ENV=test npx tsx --test test/minimal.test.ts test/modelName.test.ts test/daemonEventIntake.test.ts test/githubEventIntakeMode.test.ts test/intakeModePrerequisites.test.ts test/validateRoutingUrl.test.ts test/routingWebSocketProtocol.test.ts test/routingWebSocketIntakeService.test.ts test/routingStatusPublisher.test.ts test/prSplit/commandAuthorization.test.ts test/prSplit/operationStore.test.ts test/prSplit/intake.test.ts test/prSplit/interception.test.ts packages/api/test/statusRoutes.test.ts packages/api/test/agentRuntimeRoutes.test.ts packages/api/test/instanceAuthorization.test.ts packages/api/test/routeAuthorization.test.ts",
+    "test:unit": "NODE_ENV=test npx tsx --test test/minimal.test.ts test/modelName.test.ts test/daemonEventIntake.test.ts test/githubEventIntakeMode.test.ts test/intakeModePrerequisites.test.ts test/validateRoutingUrl.test.ts test/routingWebSocketProtocol.test.ts test/routingWebSocketIntakeService.test.ts test/routingStatusPublisher.test.ts test/partialExecution.test.ts test/prSplit/commandAuthorization.test.ts test/prSplit/operationStore.test.ts test/prSplit/intake.test.ts test/prSplit/interception.test.ts test/prSplit/analysisPlanning.test.ts packages/api/test/statusRoutes.test.ts packages/api/test/agentRuntimeRoutes.test.ts packages/api/test/instanceAuthorization.test.ts packages/api/test/routeAuthorization.test.ts",
     "test:e2e": "npx tsx --test test/e2e.test.ts",
     "test:docker": "docker-compose run --rm -e REDIS_HOST=redis -e NODE_ENV=test worker npx tsx --test test/*.test.ts test/prSplit/*.test.ts",
     "test:docker:single": "docker-compose run --rm -e REDIS_HOST=redis -e NODE_ENV=test worker npx tsx --test",

packages/core/src/agents/impl/AntigravityAgent.ts

@@ -289,7 +289,7 @@ export class AntigravityAgent implements Agent {
     }
 
     async analyze(prompt: string, options?: AnalyzeOptions): Promise<AnalysisResult> {
-        const { context, model, taskId, taskNumber, prNumber, executionType, correlationId, repository, metadata, timeoutMs, responseFormat = 'text', suppressLlmLog } = options || {};
+        const { context, model, taskId, taskNumber, prNumber, executionType, correlationId, repository, metadata, timeoutMs, signal, responseFormat = 'text', suppressLlmLog } = options || {};
         const startTime = Date.now();
         logger.info({ agentAlias: this.config.alias, promptLength: prompt.length, hasContext: !!context, requestedModel: model, taskId, executionType }, 'Running lightweight analysis via Antigravity agent...');
         const effectiveModel = model || 'antigravity-gemini-3.5-flash-medium';
@@ -302,7 +302,7 @@ export class AntigravityAgent implements Agent {
 
             const { result, usageMetrics } = await executeWithUsageTracking(
                 this.getRuntimeName(),
-                async () => executeDockerCommand('docker', dockerArgs, { timeout: timeoutMs ?? 1800000, stdinData: fullPrompt, taskId }),
+                async () => executeDockerCommand('docker', dockerArgs, { timeout: timeoutMs ?? 1800000, stdinData: fullPrompt, taskId, signal }),
                 ANALYSIS_AGENT_TANK_TIMEOUT_MS
             );
             const executionTimeMs = Date.now() - startTime;

packages/core/src/agents/impl/ClaudeAgent.ts

@@ -161,7 +161,7 @@ export class ClaudeAgent implements Agent {
 
     /** Runs a lightweight, read-only analysis for planning, summarization, and PR reviews. */
     async analyze(prompt: string, options?: AnalyzeOptions): Promise<AnalysisResult> {
-        const { context, model, taskId, taskNumber, prNumber, executionType, correlationId, repository, metadata, timeoutMs, responseFormat = 'text', reasoningLevel, useConfiguredReasoningLevel = false, suppressLlmLog } = options || {};
+        const { context, model, taskId, taskNumber, prNumber, executionType, correlationId, repository, metadata, timeoutMs, signal, responseFormat = 'text', reasoningLevel, useConfiguredReasoningLevel = false, suppressLlmLog } = options || {};
         const startTime = Date.now();
 
         logger.info({
@@ -192,7 +192,7 @@ export class ClaudeAgent implements Agent {
             const { result, usageMetrics } = await executeWithUsageTracking(
                 'claude',
                 async () => executeDockerCommand('docker', dockerArgs, {
-                    timeout: timeoutMs ?? 1800000, stdinData: analysisPrompt, taskId
+                    timeout: timeoutMs ?? 1800000, stdinData: analysisPrompt, taskId, signal
                 }),
                 ANALYSIS_AGENT_TANK_TIMEOUT_MS
             );

packages/core/src/agents/impl/CodexAgent.ts

@@ -205,7 +205,7 @@ export class CodexAgent implements Agent {
     }
 
     async analyze(prompt: string, options?: AnalyzeOptions): Promise<AnalysisResult> {
-        const { context, model, taskId, taskNumber, prNumber, executionType, correlationId, repository, metadata, timeoutMs, responseFormat = 'text', reasoningLevel, useConfiguredReasoningLevel = false, suppressLlmLog } = options || {};
+        const { context, model, taskId, taskNumber, prNumber, executionType, correlationId, repository, metadata, timeoutMs, signal, responseFormat = 'text', reasoningLevel, useConfiguredReasoningLevel = false, suppressLlmLog } = options || {};
         const startTime = Date.now();
         const effectiveModel = model || this.config.defaultModel || 'unknown';
 
@@ -232,7 +232,7 @@ export class CodexAgent implements Agent {
             const { result, usageMetrics } = await executeWithUsageTracking(
                 'codex',
                 async () => executeDockerCommand('docker', dockerArgs, {
-                    timeout: timeoutMs ?? 1800000, stdinData: analysisPrompt, taskId
+                    timeout: timeoutMs ?? 1800000, stdinData: analysisPrompt, taskId, signal
                 }),
                 ANALYSIS_AGENT_TANK_TIMEOUT_MS
             );

packages/core/src/agents/impl/OpenCodeAgent.ts

@@ -26,6 +26,13 @@ function resolveOpenCodeExecutionOutcome(
     return { success: result.exitCode === 0 && !parsedOutput.error && !terminationReason, terminationReason };
 }
 
+function buildAnalysisPrompt(prompt: string, context: string | undefined, responseFormat: 'text' | 'json'): string {
+    const suffix = responseFormat === 'json'
+        ? '\n\nCRITICAL: Do not modify any files. Do not run any commands. Return only valid JSON matching the requested schema. Do not include markdown or explanatory text.'
+        : '\n\nCRITICAL: Do not modify any files. Do not run any commands. Only provide your analysis as plain text output.';
+    return context ? `${prompt}\n\nContext:\n${context}${suffix}` : `${prompt}${suffix}`;
+}
+
 export class OpenCodeAgent implements Agent {
     readonly config: AgentConfig;
     private readonly timeoutMs: number;
@@ -121,11 +128,10 @@ export class OpenCodeAgent implements Agent {
     }
 
     async analyze(prompt: string, options?: AnalyzeOptions): Promise<AnalysisResult> {
-        const { context, model, taskId, taskNumber, prNumber, executionType, correlationId, repository, metadata, suppressLlmLog } = options || {};
+        const { context, model, taskId, taskNumber, prNumber, executionType, correlationId, repository, metadata, timeoutMs, signal, responseFormat = 'text', suppressLlmLog } = options || {};
         const startTime = Date.now();
         const effectiveModel = model || this.config.defaultModel || 'unknown';
-        const suffix = '\n\nCRITICAL: Do not modify any files. Do not run any commands. Only provide your analysis as plain text output.';
-        const analysisPrompt = context ? `${prompt}\n\nContext:\n${context}${suffix}` : `${prompt}${suffix}`;
+        const analysisPrompt = buildAnalysisPrompt(prompt, context, responseFormat);
         const analysisWorkspace = this.ensureAnalysisWorkspace();
         const analysisConfigPath = this.createAnalysisConfigSnapshot();
         const analysisDataPath = this.resolveAnalysisDataPath();
@@ -134,7 +140,7 @@ export class OpenCodeAgent implements Agent {
             const dockerArgs = await this.buildDockerArgs({ worktreePath: analysisWorkspace, githubToken: process.env.GITHUB_TOKEN || '', modelName: effectiveModel === 'unknown' ? undefined : effectiveModel, issueNumber: 0, taskId, executionType, readOnlyWorkspace: true, configPath: analysisConfigPath, dataPath: analysisDataPath });
             const { result, usageMetrics } = await executeWithUsageTracking(
                 'opencode',
-                async () => executeDockerCommand('docker', dockerArgs, { timeout: 1800000, stdinData: analysisPrompt, taskId })
+                async () => executeDockerCommand('docker', dockerArgs, { timeout: timeoutMs ?? 1800000, stdinData: analysisPrompt, taskId, signal })
             );
             const executionTimeMs = Date.now() - startTime;
             const parsedOutput = this.parseOpenCodeJsonl(result.stdout);
@@ -144,19 +150,15 @@ export class OpenCodeAgent implements Agent {
             const success = !result.timedOut && result.exitCode === 0 && !parsedOutput.error && analysisText.length > 0;
 
             const errorMsg = parsedOutput.error || result.stderr || 'No assistant text returned';
-            if (!suppressLlmLog) {
-                await this.persistAnalysisLogSafely({ executionType, modelUsed, executionTimeMs, success, error: success ? undefined : errorMsg, sessionId: parsedOutput.sessionId, taskId, correlationId, repository, metadata, taskNumber, prNumber, tokenUsage: parsedOutput.tokenUsage, usageMetrics });
-            }
+            await this.persistAnalysisLogUnlessSuppressed(suppressLlmLog, { executionType, modelUsed, executionTimeMs, success, error: success ? undefined : errorMsg, sessionId: parsedOutput.sessionId, taskId, correlationId, repository, metadata, taskNumber, prNumber, tokenUsage: parsedOutput.tokenUsage, usageMetrics });
             return success
                 ? { response: analysisText, modelUsed, executionTimeMs, success: true, sessionId: parsedOutput.sessionId, tokenUsage: parsedOutput.tokenUsage }
                 : { response: analysisText, modelUsed, executionTimeMs, success: false, error: `Analysis failed: ${errorMsg}`, tokenUsage: parsedOutput.tokenUsage };
         } catch (error) {
             const executionTimeMs = Date.now() - startTime;
             const err = error as Error;
             logger.error({ agentAlias: this.config.alias, error: err.message, executionTimeMs }, 'OpenCode lightweight analysis failed');
-            if (!suppressLlmLog) {
-                await this.persistAnalysisLogSafely({ executionType, modelUsed: effectiveModel, executionTimeMs, success: false, error: err.message, taskId, correlationId, repository, metadata, taskNumber, prNumber });
-            }
+            await this.persistAnalysisLogUnlessSuppressed(suppressLlmLog, { executionType, modelUsed: effectiveModel, executionTimeMs, success: false, error: err.message, taskId, correlationId, repository, metadata, taskNumber, prNumber });
             return { response: '', modelUsed: effectiveModel, executionTimeMs, success: false, error: err.message };
         } finally {
             this.cleanupAnalysisWorkspace(analysisWorkspace);
@@ -255,6 +257,13 @@ export class OpenCodeAgent implements Agent {
         }
     }
 
+    private async persistAnalysisLogUnlessSuppressed(
+        suppressLlmLog: boolean | undefined,
+        opts: Parameters<OpenCodeAgent['persistAnalysisLogSafely']>[0]
+    ): Promise<void> {
+        if (!suppressLlmLog) await this.persistAnalysisLogSafely(opts);
+    }
+
     private ensureAnalysisWorkspace(): string {
         const workspace = this.createAnalysisTempDir('workspace-');
         try {

packages/core/src/agents/impl/VibeAgent.ts

@@ -187,7 +187,7 @@ export class VibeAgent implements Agent {
 
     // eslint-disable-next-line complexity
     async analyze(prompt: string, options?: AnalyzeOptions): Promise<AnalysisResult> {
-        const { context, model, taskId, taskNumber, prNumber, executionType, correlationId, repository, metadata, timeoutMs, responseFormat = 'text', suppressLlmLog } = options || {};
+        const { context, model, taskId, taskNumber, prNumber, executionType, correlationId, repository, metadata, timeoutMs, signal, responseFormat = 'text', suppressLlmLog } = options || {};
         const startTime = Date.now();
         const effectiveModel = model || this.config.defaultModel;
         if (!effectiveModel) throw new NoDefaultModelConfiguredError();
@@ -226,7 +226,8 @@ export class VibeAgent implements Agent {
                 'vibe',
                 async () => executeDockerCommand('docker', dockerArgs, {
                     timeout: timeoutMs ?? parseInt(process.env.VIBE_ANALYSIS_TIMEOUT_MS || '1800000', 10),
-                    taskId
+                    taskId,
+                    signal
                 })
             );
             const executionTimeMs = Date.now() - startTime;

packages/core/src/agents/types.ts

@@ -122,6 +122,8 @@ export interface AnalyzeOptions {
     metadata?: Record<string, unknown>;
     /** Optional timeout for lightweight analysis execution. */
     timeoutMs?: number;
+    /** Cancels the underlying analysis process and its agent container. */
+    signal?: AbortSignal;
     /** Expected response format. Defaults to plain text analysis. */
     responseFormat?: 'text' | 'json';
     /** Optional per-analysis reasoning level override. */