[BUG] byte index 135 is not a char boundary; it is inside '释' (bytes 133..136)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Writing Chinese content causes Claude code to crash.
thread '<unnamed>' (4654) panicked at /rustc/ed61e7d7e242494fb7057f2657300d9e77bb4fcb/library/core/src/str/mod.rs:833:21:
byte index 135 is not a char boundary; it is inside '释' (bytes 133..136) of 解的概念。相反，"是建设社会主义的锐利武器"作为概念依存元素，其完整意义依赖于对"总路线"的阐释——
note: run with RUST_BACKTRACE=1 environment variable to display a backtrace
fatal runtime error: failed to initiate panic, error 5, aborting

What Should Happen?

Writing Chinese content causes Claude code to crash.

Error Messages/Logs

thread '<unnamed>' (4654) panicked at /rustc/ed61e7d7e242494fb7057f2657300d9e77bb4fcb/library/core/src/str/mod.rs:833:21:
byte index 135 is not a char boundary; it is inside '释' (bytes 133..136) of `解的概念。相反，"是建设社会主义的锐利武器"作为概念依存元素，其完整意义依赖于对"总路线"的阐释——`
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
fatal runtime error: failed to initiate panic, error 5, aborting

Steps to Reproduce

Claude Code Read Tool - Offset Parameter Bug Report

Test Date: 2025-12-17
Test Environment: Claude Code CLI on WSL2 (Linux 6.6.87.2)
Test File: test_unicode.txt (10 lines of Chinese content)

---

Executive Summary

The offset parameter of Claude Code's Read tool exhibits multiple bugs when processing Unicode (Chinese) text files. The issues include incorrect content retrieval, inconsistent line number formatting, and 0-indexed behavior contrary to documentation.

---

Test Configuration

Test File Structure

| Line | Content | Chars | Bytes |
|:----:|:--------|------:|------:|
| 1 | 第一行：这是中文测试内容 | 12 | 36 |
| 2 | 第二行：汉字Unicode编码测试 | 17 | 37 |
| 3 | 第三行：中英混合 English mixed 测试 | 25 | 45 |
| 4 | 第四行：特殊字符测试——破折号、引号""、符号※ | 24 | 68 |
| 5 | 第五行：多字节字符：𠀀𠀁𠀂（CJK扩展B区） | 22 | 61 |
| 6-10 | (additional Chinese text lines) | - | - |

Total: 177 characters, 448 bytes (byte/char ratio: 2.53)

---

Test Results

Test Case Matrix

| Test ID | offset | limit | Expected Line | Actual Line# | Actual Content | Prefix Format | Status |
|:-------:|:------:|:-----:|:-------------:|:------------:|:---------------|:--------------|:------:|
| T1 | 0 | 1 | Line 1 | 0 | 第一行 | 0→ | ⚠️ |
| T2 | 1 | 1 | Line 2 | 1 | 第一行 | 1→ | ❌ |
| T3 | 2 | 1 | Line 3 | 2 | 第二行 | 2→ | ⚠️ |
| T4 | 3 | 2 | Lines 4-5 | 3-4 | 第三行、第四行 | Normal | ⚠️ |
| T5 | 5 | 1 | Line 6 | 5 | 第五行 | Normal | ⚠️ |
| T6 | 10 | 1 | Line 11 | 10 | 第十行 | 10→ | ⚠️ |

---

Identified Bugs

Bug #1: Incorrect Content at offset=1 (Critical)

Description: Both offset=0 and offset=1 return the same content ("第一行"), indicating the skip logic fails at offset=1.

Expected Behavior:

• offset=0 → Line 1 content
• offset=1 → Line 2 content

Actual Behavior:

• offset=0 → Line 1 content ✓
• offset=1 → Line 1 content ✗ (should be Line 2)

Severity: Critical - breaks basic pagination functionality

---

Bug #2: Inconsistent Line Number Prefix Formatting

Description: The line number prefix format varies unexpectedly based on offset value.

Expected Format: N→ (5 spaces + line number + arrow)

Observed Anomalies:
| offset | Prefix | Issue |
|:------:|:-------|:------|
| 0 | 0→ | Normal formatting |
| 1 | 1→ | Missing leading spaces |
| 2 | 2→ | Normal formatting |
| 10 | 10→ | Missing leading spaces |

Severity: Medium - affects readability and parsing

---

Bug #3: Zero-Indexed Line Numbers

Description: Line numbers start from 0 instead of 1.

Documentation States: "Results are returned using cat -n format, with line numbers starting at 1"

Actual Behavior: offset=0 displays line number as 0, not 1

Severity: Low - inconsistent with documentation

---

Bug #4: Semantic Confusion in Offset Behavior

Description: The offset parameter behavior is inconsistent and undocumented for Unicode content.

Observations:

• offset appears to be 0-indexed
• Content mapping: offset N → Line N (not Line N+1)
• But offset=1 breaks this pattern

Potential Root Cause: The offset calculation may be mixing byte-based and line-based indexing, causing issues with multi-byte UTF-8 characters.

---

Technical Analysis

UTF-8 Encoding Impact

Chinese characters are encoded as 3 bytes in UTF-8:

| Character | Unicode | UTF-8 Bytes |
|:---------:|:-------:|:------------|
| 第 | U+7B2C | E7 AC AC |
| 一 | U+4E00 | E4 B8 80 |
| 行 | U+884C | E8 A1 8C |

If the offset parameter internally uses byte counting instead of line counting, a byte-based skip could land in the middle of a multi-byte character, causing:

• UTF-8 decoding errors
• Incorrect line boundary detection
• Content corruption

Hypothesis

The Read tool's offset implementation likely:

1. Uses 0-based indexing internally
2. Has an off-by-one error at offset=1
3. May have byte vs. line counting confusion for CJK content

---

Reproduction Steps

# Step 1: Create test file with Chinese content
cat > test_unicode.txt << 'EOF'
第一行：这是中文测试内容
第二行：汉字Unicode编码测试
第三行：中英混合 English mixed 测试
EOF

# Step 2: Test with Claude Code Read tool
# Execute in Claude Code CLI:
Read test_unicode.txt offset=0 limit=1
Read test_unicode.txt offset=1 limit=1
Read test_unicode.txt offset=2 limit=1

# Step 3: Compare results
# offset=0 and offset=1 should return different lines
# but both return "第一行" (Bug confirmed)

---

Impact Assessment

| Use Case | Impact Level | Description |
|:---------|:------------:|:------------|
| Large file pagination | High | Cannot reliably skip to specific lines |
| Incremental file reading | High | Content may be duplicated or skipped |
| Academic document editing | Medium | Affects thesis/paper editing workflows |
| Log file analysis | Medium | Line-based navigation unreliable |

---

Recommendations

For Users (Workaround)

1. Avoid using offset=1; use offset=0 or offset=2+
2. Read entire file when possible, then process locally
3. Use Bash tool with sed or awk for reliable line extraction

For Developers (Fix Suggestions)

1. Audit offset calculation logic for Unicode handling
2. Ensure line-based (not byte-based) indexing
3. Fix off-by-one error at offset=1
4. Standardize line number display format
5. Update documentation to clarify 0-indexed vs 1-indexed behavior

---

Appendix: Raw Test Output

Test T1: offset=0, limit=1

     0→第一行：这是中文测试内容

Test T2: offset=1, limit=1

1→第一行：这是中文测试内容

Test T3: offset=2, limit=1

     2→第二行：汉字Unicode编码测试

Test T6: offset=10, limit=1

10→第十行：最后一行测试结束

---

Report Generated: 2025-12-17
Reporter: Automated Test Script
Status: Awaiting Developer Review

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

Claude Code v2.0.71

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Windows Terminal

Additional Information

_No response_