
What happened
Testing on 2.1.328 with Haiku, the author found malformed rules like Bash(git with no closing bracket now trigger a warning, but a lowercase bash(git *) gives no warning and never runs—three out of three attempts failed.
Why it matters
A rule that looks correct and produces no warning is the harder failure to catch, and the warning appears as a single line on standard error while execution finishes normally, so cron jobs logging only standard output may miss it entirely.
Summaries like this, every morning.
The author had previously documented that a permission rule missing its closing bracket in Claude Code's --allowedTools passed without comment. The 2.1.326 release notes said a warning was added, so the same tests were rerun on 2.1.328 with the Haiku model, and the result was only a partial fix.
Rules with broken bracket shapes—no closing bracket, a space before the bracket, or extra characters after it—now produce a warning that even mentions checking for a missing closing bracket or space. A correctly formed Bash(git *) rule gave no warning and ran in all three attempts, and Bash(echo a,b) also worked all three times.
The gap sits with rules that are well-formed but wrong in another way. Writing the tool name in lowercase as bash(git *) draws no warning because the shape is valid, yet nothing is granted—three out of three runs failed—which the author notes is harder to notice than a malformed rule. The author also checked 31 scheduled-run logs that capture standard error in the same file and found zero warnings, and searched 12 local config files for lowercase tool names and found none. Running claude doctor from the CLI did not surface broken skills or mismatched hook conditions; the author says those appear in the interactive /doctor or /status views instead.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Home Depot says its Magic Apron AI assistant now fields millions of questions per month and is trained on prod…

Yamada Hikaru of Toyota's advanced product development division completed a "meeting quality evaluation prompt…

The article on Anthropic's 320兆円 listing flags three structural risks: fixed mega-payments to Amazon and Googl…

OpenAI says attacks pulling protected reasoning from prior requests fell from a 54.0 percent success rate on G…

Apple is hiring a senior engineer to lead its own device-controlling AI agent, competing with Muse, Dots, Duet…

Nvidia is nearing a deal to acquire AI startup Inception for at least $500 million, per three people told The…
