{"version":"2.1.300","anchor":"the-bash-tool-description-in-the-agent-claude-prompt-was-rew","canonical_anchor":"the-bash-tool-description-in-the-agent-claude-prompt-was-rew","heading":"The Bash tool description Claude reads was rewritten","tier":"notice","area":"Bash Tool","scope":"individual","heads_up":false,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.300\/e\/the-bash-tool-description-in-the-agent-claude-prompt-was-rew","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.300","markdown":"### The Bash tool description Claude reads was rewritten\n\nClaude's instructions for the Bash tool were rewritten, including how background commands and waiting work\n\n**What**\n\nThe Bash tool is how Claude runs commands on your computer, and Claude reads a description of how to use it. That description was rewritten in this release, as measured on the prompt the running client sends:\n\n- Main prompt: changed for the interactive Claude Code prompt, with and without an account, across 13 captured model setups. Another set of model setups got a different edit. The new text includes: 'Foreground `sleep` is blocked; use Monitor with an until-loop to wait on a condition.'\n\n- Agent prompt: changed in the agent-claude prompt. The new text includes: 'You will be notified when it exits - do not poll its output file while waiting.'\n\nThe longest version of the description captured under this release is 9,857 characters.\n\n**Why**\n\nThese are Claude's own instructions, so changes here can change how Claude runs commands, particularly when it waits for something or runs a command in the background. What the change does in practice is a reading of the text, not something measured.\n\n- Area: Bash Tool\n- Tier: You'll notice\n- Useful: 1\/5\n- Signal: 1\/5\n- Scope: individual\n- Heads-up: no"}