anthropics / anthropics/claude-code

[BUG] Session becomes stuck ("Determining.../Marinating...") after long session duration in large project

Ouverte
#92,594 0 commentaires 1 réaction 0 personnes assignées Voir sur GitHub
area:core bug perf:reliability platform:vscode platform:windows
Langage dominant
Python
Étoiles
145k
Forks
23.1k
Métriques de merge des PR
Métriques de PR en attente

Description

### Preflight Checklist

- [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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?

After running a long Claude Code session (~5 hours, 56% of session limit used, 81% of weekly limit used) inside a large Katalon Studio automation project, the agent got stuck showing rotating status text ("Thought for Xs" → "Bash" → "Read" → "Determining..." → "Marinating...") without producing a response or completing the turn.

Closing and reopening VS Code did not resolve it — the session remained in the same stuck state showing "Marinating..." after reopening.

### What Should Happen?

Claude Code should either complete the turn and respond, or clearly indicate it is still actively processing vs. stuck, especially in long-running sessions with large project context (multiple SKILL.md/reference files).

### Error Messages/Logs

```shell
No explicit error message. Terminal showed:
IN: find . -iname "*.tsc" -not -path "*/node_modules/*" 2>/dev/null; find . -iname "*collection*" -not -path "*/node_modules/*" -not -path "*/.git/*" 2>/dev/null
OUT: Exit code 1
```

### Steps to Reproduce

1. Open a large Katalon Studio project in VS Code with Claude Code extension (project contains custom .claude/skills/katalon/ reference files, e.g. SKILL.md, xml-templates.md, webui-keywords.md)
2. Run a long session (several hours) with multiple file reads and bash commands across the session
3. Continue prompting ("lanjutkan", "beresin") without using /compact
4. Observe that after ~5 hours of session time, the agent's status text cycles (Thought/Bash/Read/Determining/Marinating) without completing a response
5. Close and reopen VS Code — issue persists in the same session

### Claude Model

Sonnet (default)

### Is this a regression?

I don't know

### Last Working Version

_No response_

### Claude Code Version

2.1.241 (Claude Code)

### Platform

Anthropic API

### Operating System

Windows

### Terminal/Shell

PowerShell

### Additional Information

Session context: Long-running Katalon Studio test automation project (regression test suites, .ts test case files). Project uses custom Claude Code skills folder for Katalon-specific keyword/XML template references.

Guide de contribution

Aucun guide de contribution indexé pour ce dépôt

Piste de recherche

Start by trying to reproduce with Claude Code 2.1.241 in VS Code on Windows/PowerShell using a large project with .claude/skills reference files and a long session without /compact. The issue gives no code entry point or test, so first identify where session state and rotating status text are handled. Done means long-running stuck sessions either complete, recover after reopening, or clearly report their stuck/processing state.

Rédigé par le modèle d'indexation à partir du texte de l'issue.

Évaluation

Stack technique
ai-infra-agents, powershell, python, vscode
Domaine
ai, cli, developer-experience
Type d'issue
Bug
Difficulté
4/5
Temps estimé
3-5 jours
Activité
Active
Clarté
Plutôt claire
Accessibilité débutants
38/100

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.