Claude Code's isolation: 'worktree' subagents were fixed for reaching into the parent checkout four times, across releases from 2.1.203 to 2.1.222, with the documented scope shifting each time. I lay out what git -C, GIT_DIR, and destructive git commands each broke, from the official release notes.
Aug 4, 2026: the keyv worm hit 12 unrelated orgs in 3h43m, and the poisoned builds carried valid GitHub Actions provenance. Why rotating credentials first is the thing that triggers the payload.
1,560 edit pairs mined from Claude Code logs, 622 posts scanned with Sonnet, ModernBERT-ja-130m trained on a 16GB M4 Mac mini: zero false positives, 60% recall.
Claude Code sometimes emits a stray court token and raw <invoke> as plain text instead of running Read/Edit/Bash — a malformed tool call that never executes. How to tell it from a hung shell or a 'could not be parsed' error, on long Opus 4.8 sessions.
The May 19 Mini Shai-Hulud wave compromised 314 npm packages under @antv via the `atool` maintainer account. After rolling back lockfiles, payload entry points stay behind in .claude/settings.json SessionStart hooks, .vscode/tasks.json folderOpen tasks, systemd user services, and .github/workflows/codeql.yml. Concrete IoCs and the gh-token-monitor wipe ordering before rotation.
Tested MinishLab/semble on a 1595-md Astro blog: warm bm25 returns symbol definitions in 0.84s, hybrid mode loses `seasonalBanner` to the article corpus.
Bun PR #30412 merged a Zig-to-Rust rewrite into main: 6,755 commits, 1M+ lines, binaries 3–8 MB smaller. Try via bun upgrade --canary, with notes on the claude/* branch workflow and why memory safety, not speed, drove it.
Every Ghostty tab turns into 'Claude Code' when running multiple AI CLI sessions. Fix it with shell-integration-features = no-title, CLAUDE_CODE_DISABLE_TERMINAL_TITLE, and a 30-line zsh preexec/precmd/chpwd hook that tags tabs by repo name.
Went through Thariq's 20 HTML examples for practical Claude Code use. Throwaway editing UIs for ticket triage and annotated code diffs stood out. The deciding factor is whether the output's reader is human or AI.
158K lines of AI-generated C# for a Cities: Skylines II total conversion mod. CivicRAG for codebase indexing, 300+ custom Roslyn analyzers as compile-time design rules, and manual visual debugging for render bugs AI couldn't see.
Connecting a DEV article on context rot, Anthropic's 1M context guidance, and Chroma's context rot research with earlier CTX and Compresr posts. The places to watch are CLAUDE.md size, tool output accumulation, and information loss around compact—not the model name.
