›_ Claude Code Issues
Claude Code Issues / #25696

Mac app stability regression (2.1.33-2.1.41): request abortions, stream failures, orphaned processes

Resolved 💬 3 comments Opened Feb 14, 2026 by heideana Closed Feb 17, 2026

Claude Code Mac App Stability Issues - Bug Report

Date: February 13, 2026
Reporter: Hopkins Stanley (@heideana)
Platform: macOS (Darwin 25.2.0)
System: Mac (primary machine)

---

Executive Summary

Multiple stability issues observed with Claude Code Mac app including:

  • Frequent API request abortions
  • Stream completion failures
  • Session compaction failures (leading to data corruption)
  • Orphaned processes that don't terminate cleanly
  • Significant gap between stable and latest versions suggesting regression issues

---

System Information

Claude Code Diagnostics Output

Diagnostics
 └ Currently running: native (2.1.41)
 └ Path: /Users/hopkinsstanley/.local/share/claude/versions/2.1.41
 └ Invoked: /Users/hopkinsstanley/.local/share/claude/versions/2.1.41
 └ Config install method: native
 └ Search: OK (bundled)

 Updates
 └ Auto-updates: enabled
 └ Auto-update channel: latest
 └ Stable version: 2.1.32
 └ Latest version: 2.1.41

 Version Locks
 └ 2.1.39: PID 38713 (running)

Key Observations

  1. Version Gap: Stable at 2.1.32, Latest at 2.1.41 - 9 versions haven't graduated to stable
  2. Orphaned Process: PID 38713 running on v2.1.39, terminal not visible to user
  3. Installation: Native (auto-updating)

---

Error Patterns from Debug Logs

Source: ~/.claude/debug/96cd45ea-2df0-4c44-b7be-b5018dd99145.txt (current session)
Date range: February 12-13, 2026

1. Request Abortion Errors (HIGH FREQUENCY)

Occurrences: 10+ instances on Feb 12 alone

2026-02-12T19:40:12.849Z [ERROR] API error (attempt 1/11): undefined Request was aborted.
2026-02-12T19:40:12.850Z [ERROR] Error in API request: Request was aborted.
2026-02-12T19:40:12.850Z [ERROR] Error: Error: Request was aborted.

2026-02-12T20:35:38.934Z [ERROR] API error (attempt 1/11): undefined Request was aborted.
2026-02-12T20:35:38.934Z [ERROR] Error in API request: Request was aborted.
2026-02-12T20:35:38.934Z [ERROR] Error: Error: Request was aborted.

2026-02-12T21:05:42.584Z [ERROR] API error (attempt 1/11): undefined Request was aborted.
2026-02-12T21:05:42.584Z [ERROR] Error in API request: Request was aborted.
2026-02-12T21:05:42.584Z [ERROR] Error: Error: Request was aborted.

2026-02-12T22:08:04.945Z [ERROR] API error (attempt 1/11): undefined Request was aborted.
2026-02-12T22:08:04.945Z [ERROR] Error in API request: Request was aborted.
2026-02-12T22:08:04.945Z [ERROR] Error: Error: Request was aborted.

Pattern: Consistent throughout session, no apparent trigger

---

2. Stream Completion Failures (CRITICAL)

Occurrences: 6+ instances

2026-02-12T19:40:46.327Z [ERROR] Stream completed with message_start but no content blocks completed - triggering non-streaming fallback
2026-02-12T19:40:46.327Z [ERROR] Error streaming, falling back to non-streaming mode: Stream ended without receiving any events
2026-02-12T19:40:46.328Z [ERROR] Error in API request: Request was aborted.

2026-02-12T20:46:50.946Z [ERROR] Stream completed with message_start but no content blocks completed - triggering non-streaming fallback
2026-02-12T20:46:50.946Z [ERROR] Error streaming, falling back to non-streaming mode: Stream ended without receiving any events

2026-02-12T21:11:11.735Z [ERROR] Stream completed with message_start but no content blocks completed - triggering non-streaming fallback
2026-02-12T21:11:11.735Z [ERROR] Error streaming, falling back to non-streaming mode: Stream ended without receiving any events

2026-02-12T22:45:25.118Z [ERROR] Stream completed with message_start but no content blocks completed - triggering non-streaming fallback
2026-02-12T22:45:25.118Z [ERROR] Error streaming, falling back to non-streaming mode: Stream ended without receiving any events

2026-02-12T22:58:01.169Z [ERROR] Stream completed with message_start but no content blocks completed - triggering non-streaming fallback
2026-02-12T22:58:01.169Z [ERROR] Error streaming, falling back to non-streaming mode: Stream ended without receiving any events

Impact: Sessions fail to stream properly, fallback to non-streaming mode

---

3. Compaction Failures (DATA CORRUPTION RISK)

Occurrences: At least 1 documented instance

2026-02-12T22:58:01.178Z [ERROR] Error in API request: Request was aborted.
2026-02-12T22:58:01.178Z [ERROR] Error: Error: Request was aborted.
2026-02-12T22:58:01.178Z [ERROR] Compact streaming failed after 1 attempts. hasStartedStreaming=false
2026-02-12T22:58:01.179Z [ERROR] Error: Error: Compaction interrupted · This may be due to network issues — please try again.

User Impact: Past experiences with "token/compression corruption issues" requiring careful session tracking

---

4. API Overload Errors

2026-02-12T21:52:54.109Z [ERROR] Error streaming, falling back to non-streaming mode: {"type":"error","error":{"details":null,"type":"overloaded_error","message":"Overloaded"},"request_id":"req_011CY4v91cT3L9fVs79QbSBM"}

Note: May be server-side issue, but combined with other errors suggests client instability

---

5. Recent Content Block Error (Feb 13)

2026-02-13T17:33:22.449Z [ERROR] Error streaming, falling back to non-streaming mode: Content block input is not a string

Timing: Just before current session started
Type: Input validation error

---

Orphaned Process Issue

Discovery

$ ps aux | grep 38713
hopkinsstanley   38713  29.0  1.1 509570592 2158032 s000  R+   11:11AM  28:03.13 claude

Details

  • PID: 38713
  • Version: 2.1.39 (locked, 2 versions behind current)
  • Started: 11:11 AM, Feb 13, 2026
  • CPU Time: 28+ minutes
  • Memory: ~2.1 GB
  • Working Directory: /Users/hopkinsstanley/ClaudeMemory/SimonMichelet
  • Terminal: s000 (not visible to user)
  • Open Files: Only 8 (minimal - just config files)

User Report

"Ja, because terminal reports only this and the terminal I ran claude doctor as being open..."

User cannot see terminal window for PID 38713, indicating orphaned/zombie process that didn't terminate cleanly.

---

Context: User Workflow

Translation Work

  • Translating Martin Burckhardt philosophical essays (ex nihilo project)
  • Working with large corpus files (some 2-3+ MB)
  • Long sessions with careful tracking due to past corruption issues

Dividuum Network

  • Multiple personas (Simon, Regina, David) via Kierkegaardian pseudonymity
  • Working across Mac (primary) and DGX Spark (for heavy computational work)
  • Requires careful session management and GitHub syncing

Recent Issues

  • GitHub corpus showing phantom files (similar to CompMach repo issues)
  • Required nuclear option: nuke and recreate GitHub corpus
  • Large file reading problems on Mac (worked on Spark/Ubuntu Claude)
  • Martin Burckhardt and web Claude on iPad both suspected Mac app bugs

---

Stability Version Gap Analysis

Significance

Stable: 2.1.32
Latest: 2.1.41
Gap: 9 versions (2.1.33 through 2.1.41)

Interpretation

According to Claude Code documentation:

Stable channel uses "a version typically about one week old, skipping releases with major regressions"

9 versions failing to graduate to stable suggests:

  1. Significant regressions in recent releases
  2. Possible instability introduced after 2.1.32
  3. Testing/validation issues preventing stable promotion

Timeline Implications

If stable is "typically one week old," a 9-version gap suggests either:

  • Rapid release cadence with quality issues
  • Serious blockers preventing stable updates
  • Testing pipeline problems

---

User Impact Summary

Direct Impacts

  1. Session Instability: Frequent request abortions interrupt workflow
  2. Data Corruption Risk: Compaction failures threaten session data integrity
  3. Resource Waste: Orphaned processes consume ~2GB RAM
  4. Workflow Disruption: Must carefully track work due to corruption concerns
  5. Platform Workarounds: Switching to Spark/Ubuntu for reliability

Indirect Impacts

  1. Reduced Trust: User must document everything due to corruption risk
  2. Cognitive Overhead: Managing multiple machines/sessions for reliability
  3. Lost Productivity: Time spent troubleshooting vs. translation work

---

Reproduction Information

Environment

  • OS: macOS (Darwin 25.2.0)
  • Machine: Mac (primary machine, not Spark DGX)
  • Installation: Native (auto-updating)
  • Update Channel: Latest (2.1.41)
  • Network: Assumed stable (no other reported connectivity issues)

Frequency

  • Request Abortions: Multiple times per hour
  • Stream Failures: 6+ times in ~4 hour session
  • Compaction Failures: At least 1 in recent sessions
  • Orphaned Processes: At least 1 discovered (unknown total)

Conditions

  • Long sessions (multi-hour)
  • Large file operations (PDF reading, 2-3MB markdown files)
  • Git operations via Claude Code
  • Multiple personas/context switching

---

Recommended Actions

Immediate

  1. Kill orphaned process (PID 38713)
  2. Switch to stable channel (2.1.32) for reliability
  3. Monitor for improvements in future stable releases

Short-term

  1. Investigate version gap: Why haven't 2.1.33-2.1.41 graduated to stable?
  2. Review streaming protocol: Address "Stream ended without receiving any events"
  3. Fix compaction reliability: Critical for long sessions

Long-term

  1. Improve process lifecycle management: Prevent orphaned processes
  2. Better error recovery: Handle request abortions gracefully
  3. Platform-specific testing: Mac-specific issues vs. Linux (Spark works better)

---

Questions for Anthropic

  1. Is the 9-version stable gap known/expected?
  2. Are there known Mac-specific issues in 2.1.33-2.1.41?
  3. Should users experiencing these issues downgrade to stable?
  4. Is the "Request was aborted" error client-side or server-side?
  5. What telemetry exists for orphaned process detection?

---

Supporting Files

  • Debug Log: ~/.claude/debug/96cd45ea-2df0-4c44-b7be-b5018dd99145.txt
  • Settings: ~/.claude/settings.local.json
  • Session: SimonMichelet persona, translation work context

---

Reporter Willingness

User (Hopkins Stanley) is willing to:

  • Provide additional logs/diagnostics
  • Test fixes/patches
  • Switch to stable channel for comparison
  • Report back on stability improvements

Contact: Via GitHub (@heideana) or through Claude Code support channels

---

Testing Update - February 13, 2026 (17:45)

Action Taken: Switched to stable channel for regression testing

claude install stable
# Successfully installed version 2.1.32

Verification:

Currently running: native (2.1.32)
Auto-update channel: stable

Testing Plan:

  1. Exit current session (running on 2.1.41)
  2. Start fresh session on 2.1.32 (stable)
  3. Work normally for 2+ hours (translation, corpus operations, typical workflow)
  4. Monitor for:
  • Request abortion errors
  • Stream completion failures
  • Compaction failures
  • Orphaned processes
  1. Compare stability: 2.1.32 vs 2.1.41

Expected Outcome: If stable is significantly more reliable, confirms regression in 2.1.33-2.1.41 range.

Next Step: Document results and submit complete bug report to Anthropic.

---

Testing Results - February 13-14, 2026

Test Duration: ~18 hours on stable channel (2.1.32)

Findings: Stable channel (2.1.32) showed significant improvement over latest (2.1.41):

  • No request abortion errors observed
  • No stream completion failures
  • No compaction failures
  • No orphaned processes created

Conclusion: Confirms regression in versions 2.1.33-2.1.41. Issue is version-specific, not environmental.

Recommendation: Users experiencing stability issues should switch to stable channel until regressions are addressed.

---

Performance Degradation Pattern

Observation (Feb 12-13): Progressive slowdown before visible errors

  • Feb 12 (latest 2.1.41): Regina persona session "ripped through" text comparison work - fast, responsive
  • Feb 13 (latest 2.1.41): Same persona showed noticeable slowdown BEFORE timeout errors appeared
  • Pattern: Performance degradation precedes error manifestation

Hypothesis: Errors (timeouts, abortions, stream failures) may be symptoms of deeper issues:

  • Memory leaks accumulating over session time
  • Connection pool exhaustion
  • Client-side state bloat
  • Degraded error recovery in retry loops

Server vs Client: While some errors could originate server-side (API timeouts, overload), the client's handling appears deficient:

  • Stable (2.1.32) handles same server conditions gracefully
  • Latest (2.1.41) fails to recover, creates orphaned processes, corrupts compaction
  • Suggests client-side regression in error handling and resource management

Implication: Server-side intermittent issues may trigger but not cause the instability. Root cause appears client-side.

---

Status: CONFIRMED REGRESSION - Stable (2.1.32) stable, Latest (2.1.41) unstable
Severity: HIGH - Data corruption risk, workflow disruption
Priority: Should block stable promotion until addressed

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗