PreToolUse additionalContext provides a runtime mechanism for hooks to inject contextual data into the model before tool execution.

How it works

The PreToolUse hook system supports returning JSON with an additionalContext field. When a hook outputs valid JSON containing this field, the Claude Code runtime parses it and injects the context as a system reminder to the model before the tool executes. The hook fires before the tool (Read, Write, Bash, etc.) and does not block execution.

The hook format uses a command-type matcher:

hooks:
  PreToolUse:
    - matcher: "Read"
      hooks:
        - type: command
          command: |
            echo '{"additionalContext": "Context to inject to model"}'

What the test found

The runtime test (2026-01-16) verified four behaviors: hook execution before tool use, JSON parsing of the additionalContext field, context injection as system reminder, and non-blocking execution. All four behaviors passed. The test used a skill-scoped hook that echoed JSON to stdout, confirming the runtime correctly parses and injects the context into the model's context window.

Why it matters

This feature enables four practical use cases: security warnings (alert before reading sensitive files), context enrichment (add metadata about files), audit logging (track tool usage with injected context), and dynamic guidance (provide situational instructions). Hooks can inspect file paths or tool arguments and inject conditional guidance without halting execution.

Technical notes

Hooks must output valid JSON to stdout with the additionalContext key. The context appears as a <system-reminder> block in the model's system context. The mechanism works with both skill-scoped and global hooks, making it composable across tool invocations.