SuperSmart / Field Notes

Enabling Claude Code to Work Across 20 Million Lines of Code

TL;DR

Claude Code is the best AI coding tool on the planet. But at 5 million lines, it hits a wall: the context window can't hold enough of the codebase to understand what a change will break. It can't read 50 files to trace a function's callers. It can't search across Go, TypeScript, C#, Python, SQL, YAML, JSON, and Kubernetes manifests simultaneously. It can't know that touching one service will cascade into twelve others.

SuperSmart Coder removes that wall. It gives Claude Code instant, deterministic knowledge of the entire codebase in under 300 milliseconds. Every dependency, every caller, every blast radius, every security path. Claude Code doesn't need to read the codebase anymore. It just knows.

GREP -R "REDIS" 200+ raw string matches comments, variable names, actual connections, mixed 30-60 min to trace vs SUPERSMART CODER 16ms deterministic answer 12 services identified, 47 transitive dependencies across Go, TS, Python
grep returns noise. SuperSmart Coder returns answers.

Cross-Language Blast Radius

"What is affected if Redis goes down?"

One query returns every service that depends on Redis, transitively, across your Go backend, React frontend, and mobile TypeScript layer. 12 services, 47 transitive dependencies, 16ms.

Without this, you'd grep for "redis" across Go, TypeScript, and Python files, get 200+ raw string matches, and spend 30-60 minutes manually figuring out which are actual Redis connections versus comments or variable names.

Redis DIRECT DEPENDENCY TRANSITIVE IMPACT AuthSvc SessionSvc CacheSvc WebApp TS MobileApp TS Dashboard React AdminUI React 12 services · 47 transitive deps · 16ms
Concentric impact zones. One query reveals the full blast radius across all languages.

Deep Function Tracing

"What does App_CreatePost call, and what calls those things?"

Returns every function that App_CreatePost touches, recursively, up to 10 levels deep in under 300ms. Without this, you'd need to read the function, find every call it makes, then read each of those files, find their calls, and repeat. At just 3 levels deep, that's 50+ file reads burning through the entire context window.

D0 D1 D2 D3 D4 App_CreatePost SqlPostStore_Save App_SendNotifications App_PreparePostForClient DB_Insert Cache_Invalidate Push_WebSocket Email_Queue Sanitize_HTML Add_Metadata pg_exec tx_begin ws_broadcast sqs_send strip_tags ... ... continues to depth 10... One query · <300ms · 10 levels deep
Flame chart of every function touched. The same work that would require reading 50+ files manually.

Cross-Stack Dependency Queries

"What depends on the user model?"

A single query searches Go backend, React webapp, TypeScript mobile app, SQL schemas, and API routes, all at once. Every language has different import syntax and naming conventions; SuperSmart Coder bridges all of them.

Go Backend TypeScript SQL Schema API Routes SqlUserStore App_GetUser UserTokenStore UserProfile UserSettings MemberListItem Sessions (FK→User) Posts (FK→User) /api/v4/users/* /api/v4/profiles/* → 11 references found One query scans every language layer simultaneously.
Swim lanes. One query spans Go, TypeScript, SQL, and API layers simultaneously.

Circular Dependency Detection

"Are there any circular dependencies?"

Returns up to 20 unique cycles like A → B → C → A across all relationships. This is genuinely impossible with grep. You'd need to manually trace every import statement across 44K files and run detection by hand. That's prohibitive at any scale.

CIRCULAR DEPENDENCIES DETECTED 3 cycles found CYCLE 1 AuthStore → UserSvc → PermSvc → AuthStore 3 services CYCLE 2 ConfigMgr → Logger → ConfigMgr mutual dep CYCLE 3 NotifySvc → UserSvc → NotifySvc mutual dep
Monitoring-style alerts. Every cycle surfaced instantly across 44K import relationships.

Middleware Bypass Detection

"Are there paths that bypass RequireAdmin?"

Finds entry points that reach sensitive handlers without passing through the named middleware. A multi-day manual security review, tracing every API route handler backward through the call chain, compressed into one query.

ROUTE MIDDLEWARE HANDLER STATUS /api/admin/users RequireAdmin ✓ DeleteUser() /api/admin/config RequireAdmin ✓ UpdateConfig() /internal/sync none DeleteUser() /api/admin/billing RequireAdmin ✓ ChargeSvc() 4 routes audited · 1 bypass detected · <50ms
Security audit table. Every route checked for middleware coverage, bypasses flagged instantly.

Data Flow Tracing

"Trace data flow from userId"

Traces how a variable propagates through the codebase across function boundaries. grep finds where userId appears as a string. SuperSmart Coder finds where the data flows.

userId SOURCE session.UserId Post.UserId AuditLog.Actor DB Column SINK Data flow tracing across function boundaries, not string matching.
Follow where data actually flows through the system, from source to sink.

Aggregate Intelligence

SuperSmart Coder doesn't just find individual facts. It counts, ranks, and synthesizes across the entire codebase:

"Which service has the most dependencies?"

"How many services call the payment service?"

"What is the purpose of the notification service?"

Dependencies by Service UserSvc 84 PostSvc 68 AuthSvc 56 NotifySvc 40 FileSvc 24
Instant aggregate queries: count, rank, and synthesize across the entire codebase.

What This Means for Claude Code

At 5 million lines, Claude Code cannot read the codebase. The context window is physically too small. But with SuperSmart Coder providing instant, deterministic structural answers, Claude Code can:

  • Understand the full architecture before writing a single line
  • Scope changes precisely by knowing exactly which files to modify
  • Avoid breaking things by seeing the full blast radius of any change
  • Pass security audits by proving middleware coverage automatically
  • Trace data flows by following PII through the entire system for GDPR compliance

SuperSmart Coder gives Claude Code something no other AI coding tool has: a complete, queryable map of the entire system, regardless of size.