claude -p returns empty result field despite model generating response

Description

When using claude -p (print mode), the result field in the output is always empty, even though the model successfully generates a response. This makes --output-format=text return blank output, breaking programmatic/scripted usage.

Version

• Claude Code v2.1.83 (latest as of 2025-03-25)
• macOS Darwin 25.3.0

Steps to Reproduce

# Returns empty string
claude -p "Reply with exactly: hello" --output-format=text < /dev/null

# JSON output shows result is empty despite successful completion
claude -p "Reply with exactly: hello" --output-format=json < /dev/null
# => {"type":"result","subtype":"success","is_error":false,"result":"","stop_reason":"end_turn",...,"output_tokens":5}

Expected Behavior

result field should contain the assistant's response text. --output-format=text should print the response.

Actual Behavior

• result field is "" (empty string)
• output_tokens shows tokens were generated (e.g., 5 or 13)
• stop_reason is "end_turn" (normal completion)
• subtype is "success"

Evidence from Streaming

Using --verbose --output-format=stream-json, the assistant message IS visible in the stream:

{"type":"assistant","message":{"content":[{"type":"text","text":"\n\nHello! How can I help you today?"}],...}}

But the final result event still has "result":"".

Impact

This breaks any workflow that relies on capturing claude -p output programmatically, such as:

• CI/CD pipelines using Claude for code review
• Scripts that pipe Claude output to other tools
• Multi-agent setups where one CLI invokes another

Workaround

Parse the streaming JSON output (--verbose --output-format=stream-json) and extract the assistant message text manually.