How 250 tokens replaced 50,000 tokens of codebase exploration

[thegdsks]

The problem

AI coding agents spend most of their time exploring. On a 23k-line Express.js repo, Claude Code made 40+ tool calls just to understand the project structure before answering a single question.

The fix

npx stacklit init generates a ~250 token index. The agent reads one file instead of exploring the entire codebase.

Results (coming soon)

We are running controlled benchmarks on Express.js and Stacklit itself. Results will be posted here with exact numbers for:

• Tool calls (Grep, Glob, Read)
• Tokens consumed
• Wall-clock time
• Answer correctness

Try it yourself

npx stacklit init

One command. No setup. Your repo becomes AI-agent-ready.

---

Benchmark methodology and raw data will be added once testing is complete.