AIToday
AI Coding AssistantsAI Safety & AlignmentZenn AI/MLPublished: Oct 16, 2026, 22:01 JST

agent-guardrails plugin adds emergency stop to Claude Code

agent-guardrails plugin adds emergency stop to Claude Code

3 Key Points

  1. What happened

    A developer released agent-guardrails, a one-script Python standard-library plugin that guards Bash, Write, Edit, MultiEdit and NotebookEdit calls, and stops all writes if a .claude/STOP file is placed in a project.

  2. Why it matters

    This gives people who delegate write actions, such as scheduled posts or publishing products, a way to halt those actions when they need to, with the block reason shown to Claude along with a message asking it to request a settings change if the operation was intended.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

The plugin targets a common moment for people who hand work to Claude Code: wanting to stop an action that is already underway. The developer's case involved delegated tasks that write externally, such as scheduled posts and product publishing.

The design balances two risks. A broken settings file blocks rather than silently ignoring rules, while unexpected internal errors pass through and are recorded, so a script bug does not freeze the whole session. Removing the stop is deliberately left to a person, since Claude cannot delete the .claude/STOP file while writes are blocked.

The guard only covers Claude Code's tool calls. Scripts that run outside it, such as cron, CI, or a bot, need the same mechanism, including a ledger, emergency stop and caps, built into the script itself.

FAQ
How do I install agent-guardrails?
In Claude Code, run two lines: /plugin marketplace add neuralalpha-ai/agent-guardrails and /plugin install agent-guardrails@neuralalpha. Python 3 must be on your PATH; on Windows, if only python is available, the hook fails and falls on the side of not stopping.
What happens if the configuration file is broken?
A broken JSON config leads to a block, because the developer was most worried about rules silently becoming invalid due to a writing mistake. Other internal errors in the guard pass through and are only recorded, to avoid a script bug stopping the whole session.
Can Claude remove the stop file?
No. Even if asked to run /agent-guardrails:stop-agent off, Claude cannot delete the .claude/STOP file because writes are blocked; removing it is a human action.

AI news that matters for your work, delivered every morning.

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

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.

Questions and answers are published on this page.

Related Articles

Next articleBillionaires dump Apple, Microsoft, other AI stocks in Q2