← Back to the briefs
Tools

Anthropic adds workflow agent rewriting and telemetry in Claude Code v2.1.300

Anthropic released Claude Code v2.1.300, introducing mod hooks that rewrite workflow agent prompts and models, OpenTelemetry GenAI trace attributes, and strict input validation for tool hooks.

Workflow agent hooks and interface controls

Anthropic updated Claude Code to version 2.1.300, adding capabilities for extensions that manage background subagents. The agent spawn mod hook can now rewrite the prompt and target model for a workflow agent. Previously, this hook could only decline execution.

The update also provides a UI function for mods to display an agent transcript in the main interface and navigate back. Extensions can also use virtual text in prompt decorations to show interface notes that stay out of the submitted prompt text.

Input validation and execution boundaries

Anthropic revised hook security by validating input edits made during PreToolUse and PermissionRequest hooks. The engine validates these modified parameters like model input, dropping unknown fields and rejecting the invocation if the tool marks the input invalid.

Background tasks handle workflow handoffs differently. Resuming a carried workflow now requires approval through the Workflow tool instead of executing instructions directly from a handoff file. Anthropic also removed unsupported sandbox command-line flags from remote control, requiring administrators to use configuration files instead.

Telemetry and audit logs

Tracing spans now include OpenTelemetry GenAI semantic conventions, attaching provider metadata, conversation identifiers, tool names, and error classifications to runtime traces. Outbound request events also report token writes divided by cache duration.

Local logs now store full API request context, capturing system prompts, enabled tools, and supplied parameters. Users who share local transcripts can reproduce recorded interactions without missing configuration state.

Our assessment

Teams writing extensions for Claude Code should review custom hooks to ensure parameter modifications follow tool schemas, since unexpected keys will now trigger tool denials. Administrators operating shared background services should also verify that monitoring systems capture the new OpenTelemetry GenAI attributes to track subagent errors accurately.

Source: Anthropic original publication.

Original source

Read the original material

Anthropic (opens in a new tab)

The article date belongs to Agentic Horizon. The source publication date is listed separately. Vendor claims remain attributed; we have not independently tested the reported capability.

Search the publication

Find a story by title, topic, or keyword.

Press Escape to close

Ask HorizonYour site guide
Your guide to the publication

Ask Horizon.

Ask a question about the publication.

Privacy