Writing

A log check flagged 16 of 31 Claude Code turns that ended on a failing test. None ended on a false claim; it missed the one that did. Why, and 3 rules to try.

Four tests I ran on Graph Starz, a map of pictures: replaying uploads, a bulk import, scoring 16 picture models against museum records, and blind judging.

My insurer wouldn't renew over the roof's age. How I used AI, mostly Claude, to replace the policy within 45 days, pick a roofer and close two permits.

Reading all of The AI Marketing Stack changed how I diagnose this site's marketing, started a monthly read of its numbers and corrected DemandForge's prompts.

I'm testing whether my agent operator's rules and tools work better as a shared kit in project sessions: why it's worth trying and what the trial must show.

Cloud Run ends every request at its timeout, streams included, and EventSource didn't reconnect. My fix: a named ping event and a 70-second client watchdog.

I applied Jon McNeill's five-step Algorithm to six recurring Command jobs. The strongest opened 31 remediation pull requests in 30 days, and 29 merged.

I had the 38 unattended jobs in my operator's registry declare a setpoint, sensor, error signal and actuator. Sixteen aren't loops; only two close the loop.

How I use local Codex and Claude Code session logs to recover decisions, resume unfinished work, watch active tasks, and turn patterns into handoffs.

Every idea in Chip Huyen's AI Engineering, graded against my production LLM systems, plus where the book and current Anthropic guidance genuinely pull apart.

I built Karpathy's LLM wiki pattern into Command: 134 provenance-tagged pages, a gate that blocks fabricated quotes, and a human-gated publish step.

Claude Code deletes session transcripts older than 30 days by default. Raise cleanupPeriodDays in ~/.claude/settings.json to stop it; lost logs stay lost.

The key design decisions in Command, my Claude Code operator: a free deterministic watcher, per-repo config it writes, and a success criterion on every fix.

Coding agents read prose design specs as ground truth, and specs drift. chat-arch's CI guard checks each palette row as a claim; here's what it catches.

Claude conversations split across claude.ai, Cowork, and Claude Code CLI fragment your real work-streams. Site Survey is the planned fix in chat-arch.

Three URLs, one CLAUDE.md snippet, one prompt: how to load chat-arch's dark design system into any Claude Code project, and five places an LLM will miss.

My April 2026 notes on Alex Hormozi's $100M Offers, with a September update on choosing scoped technical reviews for my independent engineering work.

Two patterns from Anthropic's harness research: why agents can't grade their own work, and when to prune harness complexity as models improve.

My March 2026 notes on Alex Hormozi's $100M Leads, ranked for an independent engineer, with later updates on what changed in the plan.

Fixing three Claude Desktop failures on Windows: Cowork's 'VM service not running', an empty Google Drive connector, and 'This isn't working right now'.

One error message covers at least six root causes. Here's my diagnostic decision tree, with PowerShell commands, log locations and a fix for each.

How the Claude browser extension, Claude Code and the Google Books API turned 292 audiobooks into a reading page with covers, a pie chart and two views.

Subagents, skills, CLAUDE.md orchestration, headless runs and state management for Claude Code multi-agent systems, with 50 claims checked against the docs.

I pulled the scaffolding every MCP server I build starts with (HTTP client, Pydantic validation, 70+ tests) into a kit that turns a REST API into an MCP server.

Where Cowork conversations live on Windows, found through session manifests: transcript paths, message and tool-call records, and per-session costs.

Claude in Chrome's file_upload tool fails with CDP error -32000 on every site. The fix is one Chrome extension setting, and here's how I found it.

How I built ShopForge, an AI agent with three-tier memory, Bayesian confidence scoring and skill evolution, and what 10 days running an Etsy shop taught me.

High-frequency Cowork scheduled tasks can brick Claude Desktop on Windows and macOS. Root cause: the catch-up mechanism. Fix: rename one JSON file.

There's no official schema for Claude Desktop's Cowork session files, so I read 146 of them, diffed the fields, and documented what Anthropic stores locally.

After 1,500+ commits of AI-paired development, I use three tools with distinct roles. What stuck, what didn't, and how GPT caught a security bug in a PR.

How I found a gap in the Model Context Protocol Python SDK, implemented a fix, worked through CI failures, and got a PR ready for review in one day.

I set out to build a technical blog and three weeks later had 37 workflow documents and zero posts. Here's the research pipeline that ate the project.