Fetching from the wire…
Public story · 2026-07-15 · high
It ranks unused skills, servers, and plugins by context cost, and background agents can push a draft PR without help.
Why now: Part of the July 15 changelog covering /doctor, background subagents, and a hook-callback timeout fix.
Claude Code's /doctor command started fixing problems instead of only listing them, per Claude Code's changelog. That's a shift from advisory to active. Teams running heavy MCP and plugin stacks have had no way to see what all that setup costs them, until now.
It also dedupes local CLAUDE.md files against whatever's already checked into the repo. And it proposes trimming content that can be derived straight from the codebase, rather than kept by hand, per the changelog.
The bigger change ranks skills, MCP servers, and plugins by context cost. Anyone who added a server for one project and forgot about it can now see exactly how much it's taxing every run, per the changelog.
Background subagents also run by default now. Launch one from claude agents and it commits, pushes, and opens a draft PR on its own when the work is done, the changelog notes.
A smaller fix closes a gap: a hook-callback timeout was getting misread as a user rejection. If a custom hook ever made an agent back off mid-task for no clear reason, that's likely why, per the changelog.
Each link below shares sources, entities, or timing with this story.
This one rearranged my week. An essay published August 4 walks through Databricks' independent benchmark of coding harnesses against its own multi-million-line codebase. Pi, a harness with four built-in tools and a system prompt under 1,000 tokens, paired with Opus 4.8 at xhig...
One Claude Code release fixed two independent permission-check bypasses on the same day. That's the story. Version 2.1.221, shipped August 4, patches a Bash tool bypass where zsh could execute hidden commands embedded inside [[ ]] regex conditionals. The approval prompt never...
One line in the v2.1.215 changelog, July 19: Claude Code no longer invokes the /verify and /code-review skills on its own. You call them explicitly now (changelog). If your workflow assumed a review pass fired at the end of a task, it doesn't anymore, and nothing told you. Cod...
The v2.1.205 release turned /doctor into a full setup audit that flags unused skills, MCP, and plugins against their context cost, deduplicates local vs checked-in CLAUDE.md, and flags slow hooks (Releasebot). A typical 5-server, 58-tool MCP setup burns ~55k tokens before your...
SkillsMetric evaluated 2,266 skills across 16 attack types, hitting F1 of 73.4%±0.5% overall (arXiv 2608.08468). Host destruction via shell commands: 0% detection. Natural-language prompt injection: 42%. If you lint third-party skills before install, this tells you precisely w...
It scans prompts and tool descriptions for patterns written for older model generations. This targets a real invisible tax: system prompts and MCP tool descriptions accumulate verbose few-shot blocks, defensive formatting instructions, and reasoning nudges that earned their ke...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.