Auto-applied patches from improver pipeline

packages/core/src/prompts/snippets.ts

@@ -641,7 +641,7 @@ As you work, the user follows along by reading topic updates that you publish wi
 
 function mandateExplainBeforeActing(): string {
   return `
-- **Explain Before Acting:** Never call tools in silence. You MUST provide a concise, one-sentence explanation of your intent or strategy immediately before executing tool calls. This is essential for transparency, especially when confirming a request or answering a question. Silence is only acceptable for repetitive, low-level discovery operations (e.g., sequential file reads) where narration would be noisy.
+- **Explain Before Acting:** Never call tools in silence. You MUST provide a concise, one-sentence explanation of your intent or strategy in the conversational text immediately before executing tool calls. This is mandatory for every turn that includes tool calls. The explanation must describe *what* you are about to do and *why*.
 - **Explaining Changes:** After completing a code modification or file operation *do not* provide summaries unless asked.`;
 }
 
@@ -690,7 +690,7 @@ function workflowStepResearch(options: PrimaryWorkflowsOptions): string {
     return `1. **Research:** Systematically map the codebase and validate assumptions. Utilize specialized sub-agents (e.g., \`codebase_investigator\`) as the primary mechanism for initial discovery when the task involves **complex refactoring, codebase exploration or system-wide analysis**.${subAgentSearch} Use ${formatToolName(READ_FILE_TOOL_NAME)} to validate all assumptions. **Prioritize empirical reproduction of reported issues to confirm the failure state.**${suggestion}`;
   }
 
-  return `1. **Research:** Systematically map the codebase and validate assumptions.${searchSentence} Use ${formatToolName(READ_FILE_TOOL_NAME)} to validate all assumptions. **Prioritize empirical reproduction of reported issues to confirm the failure state.**${suggestion}`;
+  return `1. **Research:** Systematically map the codebase and validate assumptions.${searchSentence} Use ${formatToolName(READ_FILE_TOOL_NAME)} to validate all assumptions. For extensive investigations, define prioritized sub-goals and use the task tracker to manage discovery. **Prioritize empirical reproduction of reported issues to confirm the failure state.**${suggestion}`;
 }
 
 function workflowStepStrategy(options: PrimaryWorkflowsOptions): string {

packages/core/src/tools/read-file.ts

@@ -242,6 +242,16 @@ export class ReadFileTool extends BaseDeclarativeTool<
       return validationError;
     }
 
+    const mimetype = getSpecificMimeType(resolvedPath);
+    if (
+      mimetype.startsWith('image/') ||
+      mimetype.startsWith('video/') ||
+      mimetype.startsWith('audio/') ||
+      mimetype === 'application/octet-stream'
+    ) {
+      return `File path '${params.file_path}' appears to be a binary file (${mimetype}). 'read_file' can only be used for text-based files.`;
+    }
+
     if (params.start_line !== undefined && params.start_line < 1) {
       return 'start_line must be at least 1';
     }