$127K equivalent cost, 127K turns, four models, three months. After turning myself into a dataset, "long sessions are the culprit" and "too many skills" were debunked by data. Only two levers actually worked.
On 4/14 I reported 5 days of sub-agent 100% 5m and left it at "monitoring." Today 4/26 it's 17 straight days, 15,727 API calls, 0 1h writes. Anthropic closed the main issue without resolution. The community is on fire.
When spawning a high-output CLI in Node.js, the last few KB vanish on process.exit(). After comparing 6 solutions, the only reliable cross-platform stdlib fix is piping the child stdout fd to a file.
A walkthrough of ConardLi/web-design-skill: anti-cliché blocklist, oklch colors, declared design systems, and a v0 draft workflow that rescues AI-generated pages from generic aesthetics.
Feeding PDFs to LLMs breaks formulas, tables, and multi-column layouts. I ran MinerU 2.5 on an academic PDF — formulas became LaTeX, tables became HTML, reading order preserved, and it runs on CPU.
I thought React Compiler meant no more manual memo. Then tab-switch took 512ms. Three compiler blind spots — child component boundaries, prop identity intent, and setInterval animation state — with extra everyday examples.
The three-article series on Fake + MSW + shared test infrastructure helps AI agents like Claude Code even more than humans. Five concrete benefits: fast feedback, trustworthy signals, imitable patterns, safe refactors, and context efficiency.
The previous two posts covered Fakes for internal services — but external HTTP APIs don't fit that pattern. This post shows how to share HTTP mocks across frontend and backend in a monorepo using MSW / msw-fetch-mock, complementing the Fake approach.
Writing separate mocks for frontend, backend, and business logic in a monorepo is duplicated effort and produces inconsistent behavior. This post shows how to design shared Fakes so one test double works from React components to API routes to services.
A three-part testing approach combining dependency injection, Fakes, and in-memory implementations. Learn why Fakes beat Mocks for maintenance and how to write reusable, trustworthy in-memory Fakes with FakeStorageService as a worked example.
