Responses silently dropped (shown as empty) when asked to create lorem ipsum / filler text files

Description

When asking Claude Code to create a dummy file containing lorem ipsum placeholder text, the assistant's responses were silently dropped and displayed as empty (no content, no tool calls, no error). The request had to be made multiple times before any output appeared.

Steps to Reproduce

1. Open a Claude Code session in a git repository
2. Type: create a dummy file with some lorem ipsum in it for 2 paragraphs
3. Observe: response is empty / shown as (no content) — no file is created, no error is shown

Expected Behavior

Claude Code creates a file (e.g. lorem.txt) with 2 paragraphs of lorem ipsum text.

Actual Behavior

Response is silently dropped. No output, no tool calls, no error message. The request had to be retried multiple times with no success until session state changed.

Hypothesis

A content heuristic or safeguard may be misclassifying lorem ipsum (or similarly "random-looking" filler text) as suspicious content and suppressing the response entirely — rather than returning an error or declining with an explanation. Silent suppression makes this very hard to debug from the user's side.

Environment

• Claude Code CLI
• Platform: macOS (Darwin 24.6.0)
• Model: claude-sonnet-4-6

Notes

The user had to manually run touch thing.txt themselves after repeated failures. Ideally, if a safeguard is triggered, the assistant should surface a clear message rather than returning empty output.