claude-code-hidden-problem-analysis
Measured analysis of Claude Code cache bugs causing 10-20x token inflation on Max plans
- 119
- 6 mo ago
ArkNill
Full-stack engineer — SQL-native tuning across 6 DBs · Python/Java backend · AI service pipelines end-to-end
Built from ArkNill's public GitHub profile, Seoul, South Korea. Not on Take My Repo yet, so there are no posts or updates here.
Measured analysis of Claude Code cache bugs causing 10-20x token inflation on Max plans
Token-efficient browser agent for local LLMs — Playwright + accessibility tree + MarkGrab, MCP native.
Lightweight OCR + Local LLM → Schema-based Structured JSON Extraction
Unified LLM usage management — API proxy, session diagnostics, multi-CLI orchestration
Local semantic search — embedding-powered grep for files, zero external services
News monitoring pipeline — feedkit + markgrab + embgrep + diffgrab showcase
URL to screenshot with metadata. Python MCP server. Claude Vision optimized.
RSS/Atom feed collection with 449 curated feeds. Python MCP server included.
Web monitoring pipeline — track page changes, capture visual diffs, and guard against monitoring pitfalls.
Model-agnostic LLM output degeneration detector — 4-signal composite scoring, zero dependencies
Framework-agnostic agent loop detection — sliding window similarity scoring, zero dependencies
Web page change tracking + structured diffs — markgrab + snapgrab integration, MCP native
Spring Boot + Kotlin STS template without JAVA_HOME dependency
Architecture case studies — fact-checking pipelines, autonomous AI agents, and data mart automation
MarkGrab plugin for Claude Code — web content extraction to LLM-ready markdown
Declarative action policies for AI agents — composable templates for safe coding, browsing, database