Urgent.News

What's breaking now, across thousands of outlets.

AI

Bash(git push *) in a Claude Code subagent's disallowedTools took away all of Bash in 10 of 10 launches, not just git push

On Claude Code 2.1.289, a subagent whose disallowedTools said Bash(git push *) started without the Bash tool in 10 of 10 launches, so git status and echo failed along with the push, just as the docs warn. The documented alternative, a Bash(git push *) deny rule in settings, kept Bash and stopped git push origin main 3 of 3 times, but when the subagent was only handed the repository path it typed…

In a series of 10 Claude Code 2.1.289 subagent launches, the disallowedTools parameter Bash(git push *) was found to completely disable Bash, affecting not just git push, but also git status and echo. While the documented workaround using a Bash(git push *) deny rule in settings successfully prevented git push 3 out of 3 times, the subagent sometimes invoked git -C path push origin main, bypassing the restriction.

To address this issue, a custom subagent was created that could run shell commands but was expressly forbidden from pushing code. The issue was investigated by testing a lab with two throwaway git repositories containing various subagent configurations and 39 claude -p runs. The findings showed that the disallowedTools specifier in subagents effectively removes the entire tool, unlike in the main conversation or settings where it would only block specific commands.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

Is Your Phone's LLM 37x Too Slow? Check This One Compiler Flag

Is Your Phone's LLM 37x Too Slow? Check This One Compiler Flag Part 3 of a series. Part 1: found the missing -march (2.5x faster) .

  • Debug Gradle variant uses -O0 optimization level, causing 37x slower performance
  • Generation speed drops from 18.5 tokens/second on release to 0.5 on debug
  • Changing build type to Release resolves significant performance improvement

The Runtime IS the Benchmark: Endurance Metrics for Local Agents

What I Benchmarked I tested whether the runtime around the model holds up: retrieval endurance, memory scaling, sandbox containment, context behavior.

  • Runtime comparison between local and cloud models revealed memory-based hang threshold.
  • Partitioning context into 80 stores improved performance by 3.1 times over monolithic context.
  • Runtime is the key benchmark, emphasizing minimal notebook setups for optimal performance.

AI's risks are no reason for panic

“Nobody knows anything,” the screenwriter William Goldman famously said of the movie business. Much the same can be said today for artificial intelligence, a technology that’s moving fast, destination…

More from Monday 12 October →