coverme-scanner 1.10.1 → 1.10.2

package/dist/prompts/coverme-command.md

@@ -383,16 +383,34 @@ Must include at least one of:
 
 ---
 
-## PHASE 1: PARALLEL DISCOVERY
+## PHASE 1: BATCHED DISCOVERY (Memory-Efficient)
 
-Launch ALL discovery agents simultaneously.
+**CRITICAL: To prevent memory overflow, run agents in SMALL BATCHES of 3-4 agents max.**
+
+### Memory Management Rules:
+1. **Never run more than 4 agents in parallel** - prevents context overflow
+2. **Each agent returns ONLY a status message** - not full findings
+3. **Agents save findings to files** - orchestrator reads files later
+4. **Wait for batch to complete before starting next batch**
+
+### Batch Schedule:
+```
+Batch 1 (Core Security):  SEC, AUTH, API, INFRA
+Batch 2 (Specialized):    AI, BIZ, DATA, PERF
+Batch 3 (Quality):        QUAL, TEST, DEAD, PII
+Batch 4 (Deep Analysis):  REDIS, RESIL, DB, ARCH
+Batch 5 (Validation):     DESIGN, CTX, ENC, DUP
+Batch 6 (Summary):        EXEC (single agent)
+```
 
 **IMPORTANT: Each agent MUST follow these steps:**
 
 1. **Create output directory**: `mkdir -p .coverme/agents`
 2. **Scan the codebase** according to agent-specific instructions
 3. **Save findings to file**: `.coverme/agents/{AGENT_ID}.json`
-4. **Return ONLY status**: `{"status": "complete", "file": ".coverme/agents/{AGENT_ID}.json", "count": N}`
+4. **Return ONLY a short status** (not findings!): `{"status": "complete", "file": ".coverme/agents/{AGENT_ID}.json", "count": N}`
+
+**DO NOT return full findings in agent response - this causes memory overflow!**
 
 **Agent output file template:**
 ```json
@@ -1953,7 +1971,33 @@ Output as array of objects with `title` and `description` fields.
 
 ---
 
-## PHASE 7: BUILD CONSENSUS & GENERATE OUTPUT
+## PHASE 7: CONSOLIDATE AGENT RESULTS
+
+**After all batches complete, read findings from agent output files:**
+
+```bash
+# List all agent output files
+ls -la .coverme/agents/*.json 2>/dev/null
+
+# Combine all findings into one array (using jq if available, or read each file)
+cat .coverme/agents/*.json 2>/dev/null | head -100
+```
+
+**For each agent file that exists:**
+1. Read the JSON file
+2. Extract the `findings` array
+3. Merge into the master findings list
+4. Track which agents completed vs skipped
+
+**Memory-efficient consolidation:**
+- Read ONE agent file at a time
+- Extract only the findings array
+- Do NOT keep the full file content in memory
+- Build the final `scan.json` incrementally
+
+---
+
+## PHASE 8: BUILD CONSENSUS & GENERATE OUTPUT
 
 ### CRITICAL: Actually Remove False Positives!
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "coverme-scanner",
-  "version": "1.10.1",
+  "version": "1.10.2",
   "description": "AI-powered code scanner with multi-agent verification for Claude Code. One command scans everything.",
   "main": "dist/index.js",
   "files": [