Claude Code v2.1.300 adds mod prompt hooks, agent transcript views and expanded telemetry

The release adds a prompt.change mod hook, which fires when the prompt box text changes without a typed edit, such as a recalled history entry, a clear or an undo. Mods can also draw virtual text in the prompt box that never enters the draft or the submitted prompt, and $.ui.view({ agentId }) shows an agent's transcript in the main view.

Telemetry gets more detail. The OpenTelemetry api_request event now reports cache_creation_5m_tokens and cache_creation_1h_tokens, and trace spans carry GenAI semantic-convention attributes such as gen_ai.tool.name and error.type. Claude Code also adds /advisor on Claude Platform on AWS, which runs no advisor until one is selected.

Several fixes affect agent behavior. A tool whose input schema the API rejects is now sent without parameters for the rest of the conversation, instead of failing every turn with a 400 error. Agent hooks now resolve model aliases such as haiku, and background commands still running 1.5 seconds after a stop request are force-killed.

If you build mods or hooks on Claude Code, prompt.change lets you tell typed input apart from text the tool restored, and agent hooks should now use the model you configured. If your sessions hit a 400 error from an MCP tool schema, this version stops that failure, but check the new defaults for background commands and gzip-compressed requests to api.anthropic.com.