wolverine-ai 2.3.0 → 2.3.1

package/README.md

@@ -419,6 +419,24 @@ Wolverine (single process manager)
 
 ---
 
+## Cost Optimization
+
+Wolverine minimizes AI spend through 7 techniques:
+
+| Technique | What it does | Savings |
+|-----------|-------------|---------|
+| **Smart verification** | Simple errors (TypeError, ReferenceError) skip route probe — trusts syntax+boot, ErrorMonitor is safety net | Prevents $0.29 cascade |
+| **Haiku triage** | Sub-agents (explore/plan/verify/research) use cheap classifier model, only fixer uses Sonnet/Opus | 90% on sub-agent cost |
+| **Context compacting** | Every 3 agent turns, summarize history to prevent token blowup (95K→20K) | 70-80% on later turns |
+| **Cached fix patterns** | Check repair history for identical past fix before calling AI | 100% on repeat errors |
+| **Token budget caps** | Simple: 20K, moderate: 50K, complex: 100K agent budget | Caps runaway spend |
+| **Prior attempt summaries** | Pass concise "do NOT repeat" directives between iterations, not full context | Reduces baseline tokens |
+| **Backup diff context** | AI sees last known good version to revert broken code instead of patching around it | Better fix quality, fewer retries |
+
+**Result:** Simple TypeError heal drops from **$0.31 → $0.02** (15x cheaper).
+
+---
+
 ## Configuration
 
 ```

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "wolverine-ai",
-  "version": "2.3.0",
+  "version": "2.3.1",
   "description": "Self-healing Node.js server framework powered by AI. Catches crashes, diagnoses errors, generates fixes, verifies, and restarts — automatically.",
   "main": "src/index.js",
   "bin": {

package/src/brain/brain.js

@@ -112,7 +112,7 @@ const SEED_DOCS = [
     metadata: { topic: "sub-agent-tools" },
   },
   {
-    text: "Heal pipeline escalation: Iteration 1 uses fast path (CODING_MODEL, single file, cheapest). Iteration 2 uses single agent (REASONING_MODEL, multi-file, 8 turns). Iteration 3+ uses sub-agents (explore→plan→fix, 3 specialized agents with restricted tools). Each iteration gets context from previous failures. Deep research (RESEARCH_MODEL) triggers after 2+ failures.",
+    text: "Heal pipeline escalation with cost optimization: Iteration 1 uses fast path (CODING_MODEL). For simple errors (TypeError/ReferenceError/SyntaxError), verifier trusts syntax+boot and skips route probe — ErrorMonitor is safety net. This prevents false-rejection cascades that waste tokens. Iteration 2 uses single agent (REASONING_MODEL, 4 turns for simple errors, 8 for complex). Iteration 3+ uses sub-agents with Haiku for triage (explore/plan/verify/research use classifier model) and only fixer uses coding model — 90% cheaper. Token budgets capped by error complexity: simple=20K, moderate=50K, complex=100K. Context compacted every 3 agent turns to prevent token blowup (95K→20K). Prior attempt summaries passed between iterations instead of full context. Brain checked for cached fix patterns before starting AI.",
     metadata: { topic: "heal-escalation" },
   },
   {
@@ -235,6 +235,10 @@ const SEED_DOCS = [
     text: "Dependency manager skill (src/skills/deps.js): structured npm dependency analysis + repair. diagnose(errorMessage, cwd) returns {diagnosed, category, summary, fixes} — categories: missing_install, missing_package, version_conflict, outdated_api, corrupted_modules. healthReport(cwd) returns full health check: npm audit (vulnerabilities), outdated packages, peer dep conflicts, unused packages, lock file status, health score 0-100. getMigration(packageName) returns known upgrade paths: express→fastify (5.6x faster), moment→dayjs (2KB vs 70KB), request→node-fetch (deprecated), body-parser→built-in, callbacks→async/await. Agent tools: audit_deps (full health check), check_migration (upgrade paths). Heal pipeline uses diagnose() in tryOperationalFix before AI — zero tokens for dependency issues.",
     metadata: { topic: "skill-deps" },
   },
+  {
+    text: "Cost optimization: 7 techniques reduce heal cost from $0.31 to $0.02 for simple errors. (1) Verifier skips route probe for simple errors (TypeError/ReferenceError/SyntaxError) — trusts syntax+boot, ErrorMonitor is safety net. Prevents false-rejection cascades. (2) Sub-agents use Haiku (classifier model) for explore/plan/verify/research — only fixer uses Sonnet/Opus. 6 Haiku calls=$0.006 vs 6 Sonnet calls=$0.12. (3) Agent context compacted every 3 turns using compacting model — prevents 15K→95K token blowup. (4) Brain checked for cached fix patterns before AI — repeat errors cost $0. (5) Token budgets capped by error complexity: simple=20K agent budget, moderate=50K, complex=100K. Simple errors get 4 agent turns max. (6) Prior attempt summaries (not full context) passed between iterations — concise 'do NOT repeat' directives. (7) Fast path includes last known good backup code so AI can revert broken additions instead of patching around them.",
+    metadata: { topic: "cost-optimization" },
+  },
 ];
 
 class Brain {