Agent Codebase Readability & Context Window Simulator

Agent Task Target:

Cytoscape.js Codebase Traversal Map

Clean Agent-Optimized

Target Hit

Agent Active Traversal

Low-Relevance Noise

Unvisited Directory/File

CONTEXT WINDOW PRESERVED

88.1%

Standard 8,192 Token Agent Context Limit

Tokens Consumed

1,192

Traversal Hops

Hallucination Risk

4.2%

Target Relevance

91.5%

Agent Execution Traversal Log

[Ready] Select preset or task and click 'Run Traversal Simulation'

Repository Files & Token Density

Readability & Context Impact Summary

-- AUDIT REPORT SUMMARY --
Architecture: Clean Agent-Optimized
Tokens Spent: 1,192 / 8,192
Context Window Preserved: 88.1%
Hallucination Noise Risk: Low (4.2%)

Findings: High domain modularity and descriptive naming allow the agent to reach payment retry logic in 3 hops without reading monolithic junk files.

Preset context metrics and a denominator mismatch

Your browser does not support video.

Read the explanation

The page stores three preset totals: clean architecture eleven hundred ninety two tokens, monolithic six thousand nine hundred fifty, and nested four thousand eight hundred twenty. Bars use point zero six pixels per token. The clean file list sums three hundred fifty five, five hundred twenty five, two hundred twelve and one hundred to eleven hundred ninety two. These are authored fixtures rather than tokenizer measurements or observed model runs. The code declares a maximum context window of eight thousand one hundred ninety two. Subtracting clean total eleven hundred ninety two leaves seven thousand tokens, or eighty five point four five percent. Yet the displayed preserved percentage is a separate stored eighty eight point one. Monolithic remaining would be fifteen point one six percent rather than stored thirty point five. Bars compare the clean computed remainder and its stored label on the same percentage scale. The traversal replays a preset node order and logs stored relevance values. Greater than point seven is a hit; positive relevance at most point four is noise, and the middle band is a normal traversal message. Query text is exported but does not change preset scores or traversal order. Bars count the stored three and eight hops. Hallucination risk is another fixed label, not evidence of measured hallucinations; offline Cytoscape absence also limits actual graph verification.