feat: add research context to CLAUDE.md for Ralph Loop awareness

- After Ralph Wizard completes, updates case CLAUDE.md with links to all research files
- Adds /init step before starting Ralph Loop tasks to load research context
- Claude now knows where to find external resources, codebase patterns, recommendations
- Research knowledge persists across respawn cycles via CLAUDE.md

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

CLAUDE.md

@@ -16,7 +16,7 @@ When user says "COM":
 1. Increment version in BOTH `package.json` AND `CLAUDE.md`
 2. Run: `git add -A && git commit -m "chore: bump version to X.XXXX" && git push && npm run build && systemctl --user restart claudeman-web`
 
-**Version**: 0.1403 (must match `package.json`)
+**Version**: 0.1404 (must match `package.json`)
 
 ## Project Overview
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "claudeman",
-  "version": "0.1403",
+  "version": "0.1404",
   "description": "The missing control plane for Claude Code - run 20 autonomous agents with real-time monitoring and session persistence",
   "type": "module",
   "main": "dist/index.js",

src/plan-orchestrator.ts

@@ -15,8 +15,8 @@
 
 import { Session } from './session.js';
 import { ScreenManager } from './screen-manager.js';
-import { existsSync, mkdirSync, writeFileSync } from 'node:fs';
-import { join } from 'node:path';
+import { existsSync, mkdirSync, writeFileSync, readFileSync } from 'node:fs';
+import { join, dirname } from 'node:path';
 
 // ============================================================================
 // Types
@@ -919,11 +919,91 @@ ${prompt}
       writeFileSync(summaryPath, summary, 'utf-8');
 
       console.log(`[PlanOrchestrator] Saved final result to ${this.outputDir}`);
+
+      // Update the case CLAUDE.md with research context links
+      this.updateCaseClaudeMd();
     } catch (err) {
       console.error('[PlanOrchestrator] Failed to save final result:', err);
     }
   }
 
+  /**
+   * Update the case folder's CLAUDE.md with links to research and analysis files.
+   * This allows Ralph Loop to `/init` and understand the available knowledge base.
+   */
+  private updateCaseClaudeMd(): void {
+    if (!this.outputDir) return;
+
+    try {
+      // outputDir is like /path/to/case/ralph-wizard/
+      // CLAUDE.md is in the parent (case folder)
+      const caseDir = dirname(this.outputDir);
+      const claudeMdPath = join(caseDir, 'CLAUDE.md');
+
+      // Get relative path from case folder to ralph-wizard
+      const wizardRelPath = 'ralph-wizard';
+
+      // Build the research context section
+      const contextSection = `
+## Ralph Wizard Knowledge Base
+
+The following research and analysis files were generated for this task.
+Use \`/init\` to load this context, then read specific files as needed.
+
+### Research Phase
+- \`${wizardRelPath}/research/result.json\` - External resources, codebase patterns, recommendations
+- \`${wizardRelPath}/research/prompt.md\` - Research agent instructions
+
+### Analysis Phase
+- \`${wizardRelPath}/requirements/result.json\` - Requirements analysis
+- \`${wizardRelPath}/architecture/result.json\` - Architecture planning
+- \`${wizardRelPath}/testing/result.json\` - Testing strategy
+- \`${wizardRelPath}/risks/result.json\` - Risk analysis
+
+### Synthesis & Optimization
+- \`${wizardRelPath}/verification/result.json\` - Plan verification and quality scoring
+- \`${wizardRelPath}/execution-optimizer/result.json\` - Parallelization strategy
+- \`${wizardRelPath}/final-review/result.json\` - Final review and completeness check
+
+### Summary
+- \`${wizardRelPath}/summary.md\` - Human-readable plan summary
+- \`${wizardRelPath}/final-result.json\` - Complete plan with all metadata
+
+**Key Research Insights**: Check \`${wizardRelPath}/research/result.json\` for:
+- External GitHub repos and documentation links
+- Existing codebase patterns to follow
+- Technical recommendations from research
+- Potential challenges to watch for
+- Recommended tools and libraries
+`;
+
+      // Read existing CLAUDE.md or create new one
+      let existingContent = '';
+      if (existsSync(claudeMdPath)) {
+        existingContent = readFileSync(claudeMdPath, 'utf-8');
+
+        // Remove any existing Ralph Wizard section to avoid duplicates
+        const sectionStart = existingContent.indexOf('## Ralph Wizard Knowledge Base');
+        if (sectionStart !== -1) {
+          // Find the next ## heading or end of file
+          const nextSectionMatch = existingContent.slice(sectionStart + 1).match(/\n## /);
+          const sectionEnd = nextSectionMatch
+            ? sectionStart + 1 + nextSectionMatch.index!
+            : existingContent.length;
+          existingContent = existingContent.slice(0, sectionStart) + existingContent.slice(sectionEnd);
+        }
+      }
+
+      // Append the new section
+      const newContent = existingContent.trimEnd() + '\n' + contextSection;
+      writeFileSync(claudeMdPath, newContent, 'utf-8');
+
+      console.log(`[PlanOrchestrator] Updated CLAUDE.md with research context at ${claudeMdPath}`);
+    } catch (err) {
+      console.error('[PlanOrchestrator] Failed to update CLAUDE.md:', err);
+    }
+  }
+
   /**
    * Generate a human-readable summary of the plan.
    */

src/web/public/app.js

@@ -4124,7 +4124,48 @@ class ClaudemanApp {
         throw new Error(errorData.error || `Failed to write prompt file: ${writeRes.status}`);
       }
 
-      console.log('[RalphWizard] Prompt file written, sending read command to Claude...');
+      console.log('[RalphWizard] Prompt file written.');
+
+      // If plan was generated, send /init first to load CLAUDE.md with research context
+      // This gives Claude awareness of all the research files before starting tasks
+      if (hasPlan) {
+        console.log('[RalphWizard] Sending /init to load research context from CLAUDE.md...');
+        const initRes = await fetch(`/api/sessions/${sessionId}/input`, {
+          method: 'POST',
+          headers: { 'Content-Type': 'application/json' },
+          body: JSON.stringify({ input: '/init\r', useScreen: true })
+        });
+
+        if (!initRes.ok) {
+          console.warn('[RalphWizard] Failed to send /init, continuing anyway...');
+        } else {
+          // Wait for /init to complete (look for ready indicator or timeout)
+          console.log('[RalphWizard] Waiting for /init to complete...');
+          let initComplete = false;
+          for (let attempt = 0; attempt < 20; attempt++) {
+            await new Promise(r => setTimeout(r, 500));
+            try {
+              const statusRes = await fetch(`/api/sessions/${sessionId}`);
+              const statusData = await statusRes.json();
+              // /init is complete when we see the prompt indicator (❯) or status is ready
+              if (statusData.status === 'ready' || statusData.lastOutput?.includes('❯')) {
+                console.log('[RalphWizard] /init complete, Claude now has research context.');
+                initComplete = true;
+                break;
+              }
+            } catch (e) {
+              console.warn('[RalphWizard] Status check error:', e);
+            }
+          }
+          if (!initComplete) {
+            console.log('[RalphWizard] /init may still be running, proceeding anyway...');
+          }
+          // Small delay after /init
+          await new Promise(r => setTimeout(r, 1000));
+        }
+      }
+
+      console.log('[RalphWizard] Sending read command to Claude...');
 
       // Send a simple command to Claude to read the prompt file
       // This avoids all the escaping issues with long multi-line prompts