Back to List
Claude Code Bug Report: Automated Git Hard Resets Every Ten Minutes Cause Data Loss
Industry NewsClaude CodeGitAnthropic

Claude Code Bug Report: Automated Git Hard Resets Every Ten Minutes Cause Data Loss

A critical bug has been reported in Anthropic's Claude Code (version 2.1.87) where the tool automatically executes a 'git reset --hard origin/main' command every 10 minutes. This behavior, documented in GitHub issue #40710, occurs silently via programmatic Git operations rather than external binaries. The issue results in the permanent loss of all uncommitted changes to tracked files within the project repository. While untracked files and Git worktrees remain unaffected, the bug has been confirmed on macOS 15.4. Evidence from Git reflogs shows a consistent pattern of resets occurring at exact ten-minute intervals, posing a significant risk to developer productivity and data integrity.

Hacker News

Key Takeaways

  • Automated Data Loss: Claude Code version 2.1.87 performs a silent git fetch and git reset --hard origin/main every 10 minutes.
  • Scope of Impact: The bug destroys all uncommitted changes to tracked files; however, untracked files and Git worktrees are not affected.
  • Platform Specificity: The issue has been specifically reported and reproduced on macOS 15.4 using the Homebrew cask version of Claude Code.
  • Programmatic Execution: The resets are triggered via programmatic Git operations rather than spawning an external Git binary, making them harder to detect during execution.
  • Documented Evidence: Git reflogs show a precise 10-minute cadence of resets, confirming the automated nature of the bug.

In-Depth Analysis

Technical Mechanism of the Reset Bug

According to the GitHub issue report #40710, the core of the problem lies in how Claude Code interacts with the local repository's Git state. The tool appears to be programmed to synchronize with the remote origin/main branch on a fixed schedule. Specifically, it executes a git fetch origin followed by a git reset --hard origin/main. Because this is handled through programmatic Git operations—meaning it does not call the standard git command-line interface—users may not see active processes in their terminal or process monitor that indicate a reset is occurring. This silent execution is what leads to the unexpected loss of work, as any modifications to tracked files that have not been committed are overwritten by the state of the remote main branch.

Reproduction and Environment Details

The bug was identified by user johnmathews on March 29, 2026, running Claude Code version 2.1.87 on macOS 15.4 (Darwin 25.3.0, arm64). The environment utilized a compiled Bun binary installed via Homebrew. The evidence provided includes a Git reflog showing over 95 entries where the HEAD was moved to origin/main at exact intervals, such as 21:49:09, 21:59:09, and 22:09:09. This rhythmic pattern confirms that the behavior is not a one-time error but a recurring background task within the Claude Code core logic. The issue has been labeled with 'area:core', 'bug', 'data-loss', and 'has repro', indicating a high-priority technical flaw that directly impacts user data.

Industry Impact

The discovery of a data-loss bug in a major AI coding assistant like Claude Code highlights the risks associated with granting AI tools deep integration into local development environments. For the AI industry, this incident serves as a cautionary tale regarding the implementation of automated synchronization features. While the intent may be to keep the AI's context aligned with the latest remote code, the lack of user confirmation for destructive actions like a 'hard reset' violates standard version control safety protocols. This event may lead to increased scrutiny of how AI agents manage local file systems and repository states, potentially driving a shift toward more conservative, 'read-only' default behaviors or mandatory user prompts for state-changing Git commands.

Frequently Asked Questions

Question: What files are at risk when using the affected version of Claude Code?

All uncommitted changes to files currently tracked by Git are at risk of being deleted. However, the bug report specifies that untracked files survive the reset, and users working within Git worktrees are immune to this specific issue.

Question: How can I verify if my repository has been affected by this bug?

Users can check their Git reflog by running git reflog in their terminal. If the repository is affected, the log will show multiple entries labeled "reset: moving to origin/main" occurring at consistent 10-minute intervals.

Question: Is this issue limited to a specific operating system?

Based on the current bug report, the issue specifically occurs on macOS (tested on version 15.4). It is currently labeled as a platform-specific bug for macOS within the Anthropic repository.

Related News

Meituan Showcases AI Research Excellence with 32 Top Conference Papers and ACL 2026 Award
Industry News

Meituan Showcases AI Research Excellence with 32 Top Conference Papers and ACL 2026 Award

Meituan's technical team has reached a significant academic milestone in 2026, with dozens of research papers accepted by world-renowned AI conferences, including ACL, SIGIR, ICML, and KDD. To highlight these achievements, the company has curated 32 specific papers for a series of five specialized live broadcast sessions. A standout achievement in this collection is a paper recognized as an "Outstanding Paper" at ACL 2026. This initiative not only demonstrates Meituan's robust R&D capabilities in fields like natural language processing and machine learning but also emphasizes their commitment to knowledge sharing within the global technical community through detailed presentations and live replays.

Meituan Technical Team Showcases Machine Learning Research Excellence at ICML 2026 Conference
Industry News

Meituan Technical Team Showcases Machine Learning Research Excellence at ICML 2026 Conference

The Meituan Technical Team has announced a selection of academic papers accepted at the International Conference on Machine Learning (ICML) 2026. As one of the most influential international academic conferences in the field, ICML serves as a premier platform for exploring the future challenges and core issues of machine learning. Meituan's contributions focus on research that offers both significant theoretical value and practical impact. By participating in this top-tier event, the team aims to drive the development of the machine learning field and help lead future research directions through the dissemination of cutting-edge findings.

Meituan Fulfillment AI Team Showcases Frontier Agent Technology and Research Breakthroughs at ACL 2026 Special Session
Industry News

Meituan Fulfillment AI Team Showcases Frontier Agent Technology and Research Breakthroughs at ACL 2026 Special Session

The Meituan Fulfillment AI Algorithm Team recently hosted a specialized session to share their latest research findings accepted for the ACL 2026 conference. Centered on the development of a Large Language Model (LLM)-based Agent technology system, the team is focused on empowering Meituan's complex fulfillment business through self-evolving operational systems. Their research highlights significant advancements in core areas such as Continuous Pre-training (CPT), Post-training, Agentic Reinforcement Learning (RL), and multimodal understanding. With dozens of high-quality papers published in prestigious international AI conferences like ACL and EMNLP, Meituan continues to demonstrate its leadership in bridging the gap between academic innovation and industrial application, specifically within the logistics and fulfillment sectors.