[CRITICAL] Context auto-compact stuck at 0% - Tool Unusable

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/1168
2. https://github.com/anthropics/claude-code/issues/3375
3. https://github.com/anthropics/claude-code/issues/2283

If your issue is a duplicate, please close it and 👍 the existing issue instead.

_{This issue will be automatically closed as a duplicate in 3 days if there are no additional comments. To prevent auto-closure, please 👎 this comment.}

🤖 Generated with Claude Code

I started working on a smaller code and almost immediately. I got hit with " Context low (9% remaining) · Run /compact to compact & continue"... I may have to look elsewhere if this is not fixed. Thanks!

https://cap.link/qwm6s7frg8ngnnn

new feature: microcompact did not help

====================================================== Previous Conversation Compacted ======================================================
⏺ Compact summary (ctrl+r to expand) )<- COMPACT HAPPENS
⎿  Read production-criteria/FAR-140-CHECKIFEXISTS-FIX.json (885 lines)
⎿  Read production-criteria/FAR-140-FIXED-WORKFLOW.json (885 lines)
⎿  Read production-criteria/FAR-140-CONNECTION-FIX.json (885 lines)
⎿  Read production-criteria/FAR-140-FINAL-FIX.json (885 lines)
⎿  Read production-criteria/FAR-140-COMPLETE-FIX.json (885 lines)
⎿  Todo list read (7 items)

⏺ I'll now run comprehensive tests on the fixed workflow to verify all scenarios work correctly.

⏺ n8n-mcp - n8n_get_workflow (MCP)(id: "FwGMjOB0rb6OWdAR")
⎿  {
"success": true,
"data": {
… +1018 lines (ctrl+r to expand)

✳ Compacting conversation… (84s · ↓ 1.9k tokens · esc to interrupt)<-COMPACT AGAIN?

10th Aug - compaction issues happened again. Rendering it unusable.

Minutes after starting a new session, context left until auto-compact: 6% appears.

Still happening - Context left until auto-compact: 1% just 2-3 minutes of starting a session

Same issue, making Claude code currently unusable. I've gone as far as deleting ~/.claude with no change. Even on smaller code bases I get to 0% in minutes.

did the compact tool break or what? it is making it unusable. or did someone reduce the context window itself? I am having issues even when a new session is started, it runs out of context way faster then it ever did. backend context changes to the model?

this is Claudes response to this issue after doing research... pretty funny conclusion by himself. "You're absolutely right - this is a pretty significant design oversight. The research really paints a stark picture:
The Compaction System Issues:
No user control whatsoever:

Hardcoded 95% trigger threshold
No disable option for auto-compact
Black box algorithm with no transparency
Users literally getting stuck in "infinite compaction loops"

Broken failure modes:

Context showing "169%" and compacting endlessly
"102% context" loops that survive reinstallation
Users spending "90%+ of time compacting" instead of working
Tool becoming "completely unusable"

Expert consensus is damning:

"This almost never goes well on any remotely complicated change and you should basically just completely avoid compaction"
Community advice is literally "avoid the core feature entirely"

What Should Have Been Done:
✅ Configurable thresholds (compact at 70%, 80%, 90%, etc.)
✅ Disable auto-compact option
✅ Algorithm selection (aggressive vs conservative)
✅ Smart file filtering during compaction
✅ Recovery mechanisms for compaction loops
✅ Context preservation settings
The Workaround Ecosystem:
It's telling that the community has created:

Multiple Docker isolation projects to avoid the issue
Session management tools (claunch, etc.)
File-based context systems to bypass conversation memory
"Avoid compaction entirely" best practices

When users are building elaborate workarounds to avoid a core feature, that's a clear signal the feature is fundamentally broken.
The fact that paying customers on $200/month MAX plans are getting stuck in unusable compaction loops with zero recourse is... not great product design.
You've identified a legitimate gap where basic user control and configurability should have been table stakes from day one."

Thanks for the feedback here. Couple things I noticed from your pasted logs:

1. Your mcp server seems to be returning a large response
2. Post compaction it's reading a bunch of large files again. We have some file size checks here but I'm going to check again to make sure they're working as intended.

⎿  Read production-criteria/FAR-140-CHECKIFEXISTS-FIX.json (885 lines)
⎿  Read production-criteria/FAR-140-FIXED-WORKFLOW.json (885 lines)
⎿  Read production-criteria/FAR-140-CONNECTION-FIX.json (885 lines)
⎿  Read production-criteria/FAR-140-FINAL-FIX.json (885 lines)
⎿  Read production-criteria/FAR-140-COMPLETE-FIX.json (885 lines)

It's hard to tell deterministically what's going on however just from this. Next time this happens can you please submit a /bug report from Claude Code and send me the feedback id so I can look at your transcript / logs? Please make sure it's post compact + low context warning showing

was hoping the 1 M context would kick in. It did not. Stuck in compaction loop as it stands

@sid374 Hi Sid, glad to see someone from Anthropic is looking at it! │ │
│ Thank you for your report! │
│ Feedback ID: a00741c0-b96e-4499-aed5-dffa3a8734f3

Compaction kicks in close to 6% after another compaction just moments ago. Had to switch to codex.

https://share.cleanshot.com/zM9pCMX7 -> The irony of things. Compaction->Bug submitted->Compaction...

Context left until auto-compact: 10% , moments I started the CC

Using these settings have no effect:

• maxContextTokens: 200000 (increased from 150k)
• compactionThreshold: 0.95 (only compact at 95%)
• cleanupPeriodDays: 180 (6 months retention)
• maxConversationLength: 2000 (double the messages)
• aggressiveCompaction: false - Prevents aggressive trimming
• preserveCodeContext: true - Keeps code discussions
• retainSystemMessages: true - Preserves system context
• keepFileContents: false - Drops large file contents to save space
• prioritizeRecent: true - Keeps recent messages over old ones

Having this issue as well. I have completely removed Claude code and reinstalled. This happens with and without MCPs. This renders cc completely useless.

Had to use Codex which sucks but as least I am not greeted by death by compaction.. hope the 1M context will come in soon @sid374 any possible for fixes..... this affects a project I am working on.

https://streamable.com/6g1u7w My experience

@sid374 please could you raise this as critical? Lots of folks are having the same issues. I have to keep compacting and /clear does not help at all. It's EXTREMELY PAINFUL.

opencode is working hooked up to my claude pro max plan! https://opencode.ai/ @brandontan

Reduced git status from 224 → 143 lines (36% reduction)

• Removed 81 problematic files from untracked status
• Archived test artifacts that were consuming context
• Updated .gitignore to prevent future accumulation

• see whether this helps.. I doubt so as I tried claude code in a fresh repo and it happened as well

No go. Context low (11% remaining) · Run /compact to compact & continue hit right after.

Claude in Opencode is "dumber"

Context left until auto-compact: 0% remains a major issue and has a strong impact on my project deliverables

Thanks for the report and for bearing with us @brandontan. I've made some fixes and this should feel much better in tomorrow's release. Please keep an eye out and comment here if you're still running into this

no change @sid374 Context left until auto-compact: 1% ... after a couple of chats. It came back after a few chats. What are the chances of getting 1M tokens into CC?

│
│ Thank you for your report! │
│ Feedback ID: 6fba591c-18fc-4348-9ac9-cc957b2bc579 │
│

@brandontan this will be available in tomorrow's release (not released yet)

@sid374 Thanks a lot. Your guys' work does affect livelihoods.

Thanks @sid374 ! will monitor .85 before I close it

@sid374 no go... came up just a couple of chats - Context left until auto-compact: 8%

│ Thank you for your report! │
│ Feedback ID: aff1c46a-fd68-4068-af2c-d6bed9e72445

Please let me know I assist to resolve this!

@sid374 │ Thank you for your report! │
│ Feedback ID: eb09a04f-0603-46ce-969f-18857e9319fe

Also still having the same issue on .85

@sid374 I am having some data written to .claude.json - why are chat conversations being appended to this file? Is this the cause?

_},
{
"display": "farmiere project uses /Users/brtan/.claude.json so why can't this folder?",
"pastedContents": {}
},
{
"display": "[Pasted text #1 +6 lines]",
"pastedContents": {
"1": {
"id": 1,
"type": "text",
"content": "│ MCP Config locations (by scope): │\n│ • User config (available in all your projects): │\n│ • /Users/brtan/.claude.json │\n│ • Project config (shared via .mcp.json): │\n│ • /Users/brtan/Projects/pdf-multi-modal/.mcp.json (file does not exist) │\n│ • Local config (private to you in this project): │\n│ • /Users/brtan/.claude.json [project: /Users/brtan/Projects/pdf-multi-modal"_

@brandontan I'm also experiencing this. My .claude.json is 6,490,391 characters. You may be on to something.

https://github.com/anthropics/claude-code/issues/6181

@benlrichards glad I am not alone ... or else I am starting to get AI psychosis

https://github.com/anthropics/claude-code/issues/6123#issuecomment-3201870612

@brandontan I'm unsure if it's related though. I cleared all the chat history out and restart claude and still ran into the same issue.

@sid374 I am having some data written to .claude.json - why are chat conversations being appended to this file? Is this the cause?

No this should not cause this behavior.

@brandontan Looking at your transcript aff1c46a-fd68-4068-af2c-d6bed9e72445 you don't seem to have any user messages and just the compact summary which isn't taking a lot of space in your context.

• How large is your claude.md file?
• How many mcp servers do you have?
• Can you run /doctor and paste the contents here please (we've added some additional logging)
• Can you describe the behavior in more detail? How many messages / tool uses go through before you get into this state? If you could /bug right as you see the auto-compact warning pop up that would be ideal. From your last upload I'm not seeing any user messages, it's just a fresh session post compact.

@benlrichards I looked at your transcript and it had 26 tool uses with some very large file edits and 2 user messages. These consumed a bunch of tokens that led to the auto-compaction. I don't see any issues with early compaction based on what I'm seeing here. Can you also describe in detail what you're seeing and what you expect to see?

@sid374

1. How large is your claude.md file? 41KB
2. How many mcp servers do you have? 12
3. Diagnostics

└ Currently running: unknown (1.0.85)
└ Path: /opt/homebrew/Cellar/node/24.3.0/bin/node
└ Invoked: /Users/brtan/.bun/bin/claude
└ Config install method: unknown
└ Auto-updates enabled: true
└ Search: OK (vendor)

Memory Files · 1.1k tokens
└ Project (/Users/brtan/CLAUDE.md): 1.1k tokens

MCP Tools · 10.9k tokens
└ mcp__n8n-mcp__tools_documentation (n8n-mcp): 0k tokens
└ mcp__context7__resolve-library-id (context7): 0.3k tokens
└ mcp__sequential-thinking__sequentialthinking (sequential-thinking): 0.7k tokens
└ mcp__n8n-mcp__list_nodes (n8n-mcp): 0.1k tokens
└ mcp__context7__get-library-docs (context7): 0.1k tokens
└ mcp__n8n-mcp__get_node_info (n8n-mcp): 0k tokens
└ mcp__n8n-mcp__search_nodes (n8n-mcp): 0k tokens
└ mcp__n8n-mcp__list_ai_tools (n8n-mcp): 0k tokens
└ mcp__n8n-mcp__get_node_documentation (n8n-mcp): 0k tokens
└ mcp__filesystem__read_file (filesystem): 0k tokens
└ mcp__playwright__browser_close (playwright): 0k tokens
└ mcp__linear__list_comments (linear): 0k tokens
└ mcp__n8n-mcp__get_database_statistics (n8n-mcp): 0k tokens
└ mcp__filesystem__read_text_file (filesystem): 0.1k tokens
└ mcp__playwright__browser_resize (playwright): 0k tokens
└ mcp__linear__create_comment (linear): 0k tokens
└ mcp__n8n-mcp__get_node_essentials (n8n-mcp): 0k tokens
└ mcp__filesystem__read_media_file (filesystem): 0k tokens
└ mcp__playwright__browser_console_messages (playwright): 0k tokens
└ mcp__linear__list_cycles (linear): 0k tokens
└ mcp__n8n-mcp__search_node_properties (n8n-mcp): 0k tokens
└ mcp__filesystem__read_multiple_files (filesystem): 0.1k tokens
└ mcp__playwright__browser_handle_dialog (playwright): 0k tokens
└ mcp__linear__get_document (linear): 0k tokens
└ mcp__n8n-mcp__get_node_for_task (n8n-mcp): 0k tokens
└ mcp__filesystem__write_file (filesystem): 0.1k tokens
└ mcp__playwright__browser_evaluate (playwright): 0k tokens
└ mcp__linear__list_documents (linear): 0k tokens
└ mcp__n8n-mcp__list_tasks (n8n-mcp): 0k tokens
└ mcp__filesystem__edit_file (filesystem): 0k tokens
└ mcp__playwright__browser_file_upload (playwright): 0k tokens
└ mcp__linear__get_issue (linear): 0k tokens
└ mcp__github__create_or_update_file (github): 0k tokens
└ mcp__postgres__query (postgres): 0k tokens
└ mcp__docker__run_command (docker): 0k tokens
└ mcp__n8n-mcp__validate_node_operation (n8n-mcp): 0k tokens
└ mcp__filesystem__create_directory (filesystem): 0.1k tokens
└ mcp__playwright__browser_install (playwright): 0k tokens
└ mcp__linear__list_issues (linear): 0k tokens
└ mcp__github__search_repositories (github): 0k tokens
└ ListMcpResourcesTool (postgres): 0.1k tokens
└ mcp__n8n-mcp__validate_node_minimal (n8n-mcp): 0k tokens
└ mcp__filesystem__list_directory (filesystem): 0.1k tokens
└ mcp__playwright__browser_press_key (playwright): 0k tokens
└ mcp__linear__create_issue (linear): 0k tokens
└ mcp__github__create_repository (github): 0k tokens
└ ReadMcpResourceTool (postgres): 0.1k tokens
└ mcp__n8n-mcp__get_property_dependencies (n8n-mcp): 0k tokens
└ mcp__filesystem__list_directory_with_sizes (filesystem): 0.1k tokens
└ mcp__playwright__browser_type (playwright): 0k tokens
└ mcp__linear__update_issue (linear): 0k tokens
└ mcp__github__get_file_contents (github): 0k tokens
└ mcp__n8n-mcp__get_node_as_tool_info (n8n-mcp): 0k tokens
└ mcp__filesystem__directory_tree (filesystem): 0.1k tokens
└ mcp__playwright__browser_navigate (playwright): 0k tokens
└ mcp__linear__list_issue_statuses (linear): 0k tokens
└ mcp__github__push_files (github): 0k tokens
└ mcp__n8n-mcp__list_node_templates (n8n-mcp): 0k tokens
└ mcp__filesystem__move_file (filesystem): 0.1k tokens
└ mcp__playwright__browser_navigate_back (playwright): 0k tokens
└ mcp__linear__get_issue_status (linear): 0k tokens
└ mcp__github__create_issue (github): 0k tokens
└ mcp__supabase__create_branch (supabase): 0.1k tokens
└ mcp__zen__chat (zen): 0.2k tokens
└ mcp__kg-memory__create_entities (kg-memory): 0k tokens
└ mcp__n8n-mcp__get_template (n8n-mcp): 0k tokens
└ mcp__filesystem__search_files (filesystem): 0.1k tokens
└ mcp__playwright__browser_navigate_forward (playwright): 0k tokens
└ mcp__linear__list_my_issues (linear): 0k tokens
└ mcp__github__create_pull_request (github): 0k tokens
└ mcp__supabase__list_branches (supabase): 0k tokens
└ mcp__zen__thinkdeep (zen): 0.2k tokens
└ mcp__kg-memory__create_relations (kg-memory): 0k tokens
└ mcp__n8n-mcp__search_templates (n8n-mcp): 0k tokens
└ mcp__filesystem__get_file_info (filesystem): 0.1k tokens
└ mcp__playwright__browser_network_requests (playwright): 0k tokens
└ mcp__linear__list_issue_labels (linear): 0k tokens
└ mcp__github__fork_repository (github): 0k tokens
└ mcp__supabase__delete_branch (supabase): 0k tokens
└ mcp__zen__planner (zen): 0.3k tokens
└ mcp__kg-memory__add_observations (kg-memory): 0k tokens
└ mcp__n8n-mcp__get_templates_for_task (n8n-mcp): 0k tokens
└ mcp__filesystem__list_allowed_directories (filesystem): 0.1k tokens
└ mcp__playwright__browser_take_screenshot (playwright): 0k tokens
└ mcp__linear__create_issue_label (linear): 0k tokens
└ mcp__github__create_branch (github): 0k tokens
└ mcp__supabase__merge_branch (supabase): 0k tokens
└ mcp__zen__consensus (zen): 0.3k tokens
└ mcp__kg-memory__delete_entities (kg-memory): 0k tokens
└ mcp__n8n-mcp__validate_workflow (n8n-mcp): 0k tokens
└ mcp__playwright__browser_snapshot (playwright): 0k tokens
└ mcp__linear__list_projects (linear): 0k tokens
└ mcp__github__list_commits (github): 0k tokens
└ mcp__supabase__reset_branch (supabase): 0k tokens
└ mcp__zen__codereview (zen): 0.3k tokens
└ mcp__kg-memory__delete_observations (kg-memory): 0k tokens
└ mcp__n8n-mcp__validate_workflow_connections (n8n-mcp): 0k tokens
└ mcp__playwright__browser_click (playwright): 0k tokens
└ mcp__linear__get_project (linear): 0k tokens
└ mcp__github__list_issues (github): 0k tokens
└ mcp__supabase__rebase_branch (supabase): 0k tokens
└ mcp__zen__precommit (zen): 0.3k tokens
└ mcp__kg-memory__delete_relations (kg-memory): 0k tokens
└ mcp__n8n-mcp__validate_workflow_expressions (n8n-mcp): 0k tokens
└ mcp__playwright__browser_drag (playwright): 0k tokens
└ mcp__linear__create_project (linear): 0k tokens
└ mcp__github__update_issue (github): 0k tokens
└ mcp__supabase__list_tables (supabase): 0k tokens
└ mcp__zen__debug (zen): 0.3k tokens
└ mcp__kg-memory__read_graph (kg-memory): 0k tokens
└ mcp__n8n-mcp__n8n_create_workflow (n8n-mcp): 0k tokens
└ mcp__playwright__browser_hover (playwright): 0k tokens
└ mcp__linear__update_project (linear): 0k tokens
└ mcp__github__add_issue_comment (github): 0k tokens
└ mcp__supabase__list_extensions (supabase): 0k tokens
└ mcp__zen__secaudit (zen): 0.3k tokens
└ mcp__kg-memory__search_nodes (kg-memory): 0k tokens
└ mcp__n8n-mcp__n8n_get_workflow (n8n-mcp): 0k tokens
└ mcp__playwright__browser_select_option (playwright): 0k tokens
└ mcp__linear__list_project_labels (linear): 0k tokens
└ mcp__github__search_code (github): 0k tokens
└ mcp__supabase__list_migrations (supabase): 0k tokens
└ mcp__zen__docgen (zen): 0.3k tokens
└ mcp__kg-memory__open_nodes (kg-memory): 0k tokens
└ mcp__n8n-mcp__n8n_get_workflow_details (n8n-mcp): 0k tokens
└ mcp__playwright__browser_tab_list (playwright): 0k tokens
└ mcp__linear__list_teams (linear): 0k tokens
└ mcp__github__search_issues (github): 0k tokens
└ mcp__supabase__apply_migration (supabase): 0k tokens
└ mcp__zen__analyze (zen): 0.3k tokens
└ mcp__n8n-mcp__n8n_get_workflow_structure (n8n-mcp): 0k tokens
└ mcp__playwright__browser_tab_new (playwright): 0k tokens
└ mcp__linear__get_team (linear): 0k tokens
└ mcp__github__search_users (github): 0k tokens
└ mcp__supabase__execute_sql (supabase): 0.1k tokens
└ mcp__zen__refactor (zen): 0.3k tokens
└ mcp__n8n-mcp__n8n_get_workflow_minimal (n8n-mcp): 0k tokens
└ mcp__playwright__browser_tab_select (playwright): 0k tokens
└ mcp__linear__list_users (linear): 0k tokens
└ mcp__github__get_issue (github): 0k tokens
└ mcp__supabase__get_logs (supabase): 0.1k tokens
└ mcp__zen__tracer (zen): 0.4k tokens
└ mcp__n8n-mcp__n8n_update_full_workflow (n8n-mcp): 0k tokens
└ mcp__playwright__browser_tab_close (playwright): 0k tokens
└ mcp__linear__get_user (linear): 0k tokens
└ mcp__github__create_release (github): 0k tokens
└ mcp__supabase__get_advisors (supabase): 0.1k tokens
└ mcp__zen__testgen (zen): 0.3k tokens
└ mcp__n8n-mcp__n8n_update_partial_workflow (n8n-mcp): 0.1k tokens
└ mcp__playwright__browser_wait_for (playwright): 0k tokens
└ mcp__linear__search_documentation (linear): 0k tokens
└ mcp__github__list_releases (github): 0k tokens
└ mcp__supabase__get_project_url (supabase): 0k tokens
└ mcp__zen__challenge (zen): 0.6k tokens
└ mcp__n8n-mcp__n8n_delete_workflow (n8n-mcp): 0k tokens
└ mcp__github__delete_release (github): 0k tokens
└ mcp__supabase__get_anon_key (supabase): 0k tokens
└ mcp__zen__listmodels (zen): 0.1k tokens
└ mcp__n8n-mcp__n8n_list_workflows (n8n-mcp): 0k tokens
└ mcp__github__get_release_asset (github): 0k tokens
└ mcp__supabase__generate_typescript_types (supabase): 0k tokens
└ mcp__zen__version (zen): 0k tokens
└ mcp__n8n-mcp__n8n_validate_workflow (n8n-mcp): 0k tokens
└ mcp__github__upload_release_asset (github): 0k tokens
└ mcp__supabase__search_docs (supabase): 1.6k tokens
└ mcp__n8n-mcp__n8n_trigger_webhook_workflow (n8n-mcp): 0k tokens
└ mcp__github__create_tag (github): 0k tokens
└ mcp__supabase__list_edge_functions (supabase): 0k tokens
└ mcp__n8n-mcp__n8n_get_execution (n8n-mcp): 0k tokens
└ mcp__github__create_pull_request_review (github): 0k tokens
└ mcp__supabase__deploy_edge_function (supabase): 0.1k tokens
└ mcp__n8n-mcp__n8n_list_executions (n8n-mcp): 0k tokens
└ mcp__github__submit_pull_request_review (github): 0k tokens
└ mcp__n8n-mcp__n8n_delete_execution (n8n-mcp): 0k tokens
└ mcp__github__dismiss_pull_request_review (github): 0k tokens
└ mcp__n8n-mcp__n8n_health_check (n8n-mcp): 0k tokens
└ mcp__github__create_commit_status (github): 0k tokens
└ mcp__n8n-mcp__n8n_list_available_tools (n8n-mcp): 0k tokens
└ mcp__github__get_commit_statuses (github): 0k tokens
└ mcp__n8n-mcp__n8n_diagnostic (n8n-mcp): 0k tokens
└ mcp__github__get_combined_status (github): 0k tokens
└ mcp__github__get_rate_limit (github): 0k tokens
└ mcp__github__create_gist (github): 0k tokens
└ mcp__github__list_gists (github): 0k tokens
└ mcp__github__get_gist (github): 0k tokens
└ mcp__github__list_projects (github): 0k tokens
└ mcp__github__create_project (github): 0k tokens
└ mcp__github__list_project_columns (github): 0k tokens
└ mcp__github__create_project_column (github): 0k tokens
└ mcp__github__create_project_card (github): 0k tokens
└ mcp__github__list_org_packages (github): 0k tokens
└ mcp__github__list_user_packages (github): 0k tokens
└ mcp__github__list_repo_packages (github): 0k tokens
└ mcp__github__get_org_package (github): 0k tokens
└ mcp__github__get_user_package (github): 0k tokens
└ mcp__github__get_repo_package (github): 0k tokens
└ mcp__github__get_pull_request_diff (github): 0k tokens

Agent Configurations
└ Loaded custom agents: 18
└ frontend-developer (Project, gitignored)
└ sa-checker (Project, gitignored)
└ documentation-specialist (Project, gitignored)
└ n8n-devops-engineer (Project, gitignored)
└ business-analyst-pm (Project, gitignored)
└ data-platform-engineer (Project, gitignored)
└ ai-system-architect (Project, gitignored)
└ ai-infrastructure-engineer (Project, gitignored)
└ validation-agent (Project, gitignored)
└ context-manager (Project, gitignored)
└ comprehensive-qa-engineer (Project, gitignored)
└ ai-security-specialist (Project, gitignored)
└ agent-organizer (Project, gitignored)
└ analytics-dashboard-builder (Project, gitignored)
└ linear-project-manager (Project, gitignored)
└ solutions-architect (Project, gitignored)
└ n8n-workflow-engineer (Project, gitignored)
└ production-monitoring-specialist (Project, gitignored)

1. "If you could /bug right as you see the auto-compact warning pop up that would be idea

Thank you for your report! │ Feedback ID: 4f196272-443e-4190-bac7-ef6f5ec17beb

@sid374 appreciate if you can resolve this asap. My project deliverables are due soon. Thanks!

@brandontan looking at your latest transcript it seems like you're starting off with 100k tokens. This is most likely your agent files having really long description fields. To confirm can you run /context and paste the output here please?

@sid374 Please see this bug report: │ Thank you for your report! │
│ Feedback ID: 75989b27-8385-4eb0-8e9e-7d4451771f34

I have the low context/auto-compact warning saying 12% remains, but the /context shows only 22% used.

<img width="698" height="202" alt="Image" src="https://github.com/user-attachments/assets/6f6ba5aa-7194-4aa6-9840-9464e1e05798" />

<img width="367" height="96" alt="Image" src="https://github.com/user-attachments/assets/87614c56-bd7f-40a6-8d3e-65956b47c78c" />

@sid374 Just submitted another one: Thank you for your report! │
│ Feedback ID: 09062352-386c-496e-8fbe-aec265b08b80

This time I caught it at 1% left before compact with /context showing 23% used (77% remaining)

<img width="363" height="100" alt="Image" src="https://github.com/user-attachments/assets/bdd3be3f-9a47-495a-8306-9ff64e36063b" />
<img width="641" height="199" alt="Image" src="https://github.com/user-attachments/assets/b61f87a8-3d58-4957-ac04-307f4fdb6924" />

This started in the past 24 hours for me. Makes the tool completely unusable.

@sid374 please assist. Thanks!

/context

⎿  ⛁ ⛁ ⛀ ⛁ ⛁ ⛀ ⛁ ⛁ ⛁ ⛁
⛁ ⛀ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ Context Usage
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ claude-opus-4-1-20250805 • 69k/200k tokens (34%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛶ ⛶ ⛶ ⛶ ⛶
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ System prompt: 4.3k tokens (2.2%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ System tools: 1.4k tokens (0.7%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ MCP tools: 5.2k tokens (2.6%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Memory files: 11.3k tokens (5.7%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Custom agents: 46.5k tokens (23.3%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ Free space: 131.3k (65.6%)

I will remove the agents and see how it goes

 ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁
⛁ ⛁ ⛁ ⛁ ⛀ ⛀ ⛶ ⛶ ⛶ ⛶ Context Usage
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ claude-opus-4-1-20250805 • 30k/200k tokens (15%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ System prompt: 4.3k tokens (2.2%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ System tools: 1.5k tokens (0.8%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ MCP tools: 12.7k tokens (6.4%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Memory files: 11.3k tokens (5.7%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Messages: 190 tokens (0.1%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ Free space: 169.9k (84.9%)

• will monitor

<img width="925" height="347" alt="Image" src="https://github.com/user-attachments/assets/b06e5448-bc8d-45eb-b992-2ab5574dcd34" /> What on earth is going on?

No change:

/context

⎿  ⛁ ⛁ ⛁ ⛀ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ Context Usage
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ claude-opus-4-1-20250805 • 55k/200k tokens (28%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛀ ⛁ ⛶ ⛶
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ System prompt: 4.3k tokens (2.2%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ System tools: 3.1k tokens (1.5%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ MCP tools: 13.4k tokens (6.7%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Tool use & results: 20.9k tokens (10.5%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Memory files: 11.3k tokens (5.7%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Messages: 2.2k tokens (1.1%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ Free space: 144.8k (72.4%)

Context left until auto-compact: 8%

Claude Code has basically been useless for two weeks. Tasks that used to take 5 minutes take two hours. This auto-compact bug (or intentional feature to reduce tokens) has ruined Claude and made me switch to other tools.

Does anybody know if this is going to bee fixed, or is this the experience we should expect from now on?

@sid374 Have these bug reports and screenshots helped you identify the issue?

We had an issue with /context that was undercounting mcp tools + built in tools. This is fixed in 1.0.88.

If you're still running into context / compact issues, you're most likely running into issues where something is inadvertently blowing up your context. The main ways to do this are:

• large claude.md files
• large cumulative size of agents description fields
• large cumulative size of mcp tools description fields
• hooks injecting a large amount of context

@brandontan @benlrichards @arsn3152 I'd love for you guys to try this to isolate your issues:

1. Upgrade to 1.0.88
2. Start Claude Code without any mcp servers or agents
3. Try to repro the issue

Starting 1.0.89 (next version, currently unreleased at time of writing the comment) we have additional debug logs when you do /bug which will help figure out what's happening here.

Thanks... I will monitor @sid374

Thank you for your report! │
│ Feedback ID: 01202fb0-c0af-4129-8993-8f582564e098

Autocompaction is still coming on faster.
Context left until auto-compact: 3%

│
│ Thank you for your report! │
│ Feedback ID: 2e3b5f06-1ced-4cdd-9934-1e661e832812

Context left until auto-compact: 5%

It seems worse ... I am getting back to back compaction now. Please see pic.

<img width="2958" height="944" alt="Image" src="https://github.com/user-attachments/assets/807ccffd-5090-4ddf-ae18-c15f4e01413d" />

2.5 hrs = compaction ;-)

<img width="830" height="108" alt="Image" src="https://github.com/user-attachments/assets/46cf4760-564f-4526-a7dd-b38a181f2e77" />

Starting 1.0.89 - I am on 1.089 .. now .. how can I provide more info to stop this madness?

│ Thank you for your report! │
│ Feedback ID: 15612b1d-5ed8-48f7-8777-a7bcadeee0a4

1 min after last compaction , Context left until auto-compact: 2% kicks in

1.0.89 is worse.. it's fanatical about auto-compaction!

<img width="2962" height="884" alt="Image" src="https://github.com/user-attachments/assets/b3c1f019-b337-4515-905f-3a508fbad37b" />

Interesting observation:

I did /clear and /compact twice

1. Initially it was 95% and
2. now it's 100%

** Context Usage
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ claude-opus-4-1-20250805 • 200k/200k tokens (100%)
Bizzate

hmmm.. had to use OSS .. despited just re-subscribed to Max. Sad state of affairs

For the nth time, I nuked my claude installation. Fingers crossed.

with MCPs stripped, I have more context obviously and yet not workable as MCPs are crucial. 200K is too little.

mine's not stuck at 0% but it's auto-compacting way too frequently. max btw

<img width="1448" height="520" alt="Image" src="https://github.com/user-attachments/assets/2c161c85-a7ee-43fb-aa3e-450bd5eb6123" />

Bare boned. Not practical!

⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛀ ⛁ Context Usage
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ claude-opus-4-1-20250805 • 128k/200k tokens (64%)
⛀ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ System prompt: 17.2k tokens (8.6%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ System tools: 13.7k tokens (6.8%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ MCP tools: 9.3k tokens (4.6%)
⛁ ⛁ ⛁ ⛁ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Custom agents: 1.9k tokens (1.0%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Memory files: 4.9k tokens (2.5%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Messages: 80.7k tokens (40.4%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ Free space: 72.3k (36.1%)

and yet, I got it hit with auto-compaction hell : Context left until auto-compact: 12%

│ Submit Bug Report │
│ │
│ Thank you for your report! │
│ Feedback ID: a1fa7678-32d2-4e5e-85b3-d3e44051a26b │
│

FYA, @sid374

Version:1.0.90

/context

⎿  ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛀ ⛁ Context Usage
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ claude-opus-4-1-20250805 • 188k/200k tokens (94%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ System prompt: 17.2k tokens (8.6%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ System tools: 13.7k tokens (6.8%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ MCP tools: 95.9k tokens (47.9%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ Custom agents: 1.9k tokens (1.0%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ Memory files: 6.4k tokens (3.2%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ Messages: 52.5k tokens (26.3%)
⛁ ⛁ ⛁ ⛁ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ Free space: 12.4k (6.2%)

Context left until auto-compact: 0% <- Should not be the case

Thank you for your report! │
│ Feedback ID: 397f7d1f-d3d5-433a-b6ce-ae2d6160eb0e

Found a "fix." Not sure how detrimental it is to context awareness, but deleting/renaming the settings.local file in the .claude folder fixed the context issues for me. I am guessing the file just got too large for Claude to be able to effectively read because of all the bash commands it stored away that weren't reusable in most cases.

if you want to YOLO it, you can use "claude --dangerously-skip-permissions" to bypass permission prompts (the ones that when saved to the project, fill up the setting.local.json file and ultimately cause the context issue)

But that is absolute last resort and very dangerous to do. Hence the flag name....

All you have to do is wait, the UI doesn't update - sometimes it can take 5 minutes to compact depending on load.

All you have to do is wait, the UI doesn't update - sometimes it can take 5 minutes to compact depending on load.

I don't think that's the problem being outlined here. There's a problem where the settings.local.json file will slowly consume all of Claude's context limit as you choose to let Claude do actions without prompting again.

Which means that Claude will read a file, then immediately start compacting again. It's not that it's getting stuck doing a single compact action.

I ran into this myself Saturday night. Sunday afternoon I came across the workaround of just renaming the setting file with all of the stored actions. You could most likely go in and remove individual actions to decrease the size of the file, which also means you don't have to go in and re-approve actions that you had done previously. (This is probably the best course of action, but it is the most tedious one)

@brandontan Looking at your transcripts and feedback results again it seems like your MCP tools are giant. You've got:

• 160+ MCP tools registered - this fills up your context by 80-90k tokens
• You also have a lot of agents defined that also eat up a bunch of space
• Claude Code currently auto-compacts around ~155k tokens so that we have enough buffer for output tokens and the summary response

What you are seeing is expected behavior.

The only thing that still seems weird to me is the consecutive auto-compacts. If you're able to send me a /bug report for that with the feedback_id I can look into that. My recommendation here is to be selective about which mcp tools you are polluting your context with. Looking at your previous bug reports, I saw maybe 1 mcp tool being called in your session (but 160+ tools registered).

@Codewow the settings.local.json should have nothing to do with auto-compaction. If you're able to share your feedback_id after running /bug and paste your settings.local.json file here - I can help validate. (please make sure that you are pre-autocompact with you run /bug)

@Codewow the settings.local.json should have nothing to do with auto-compaction. If you're able to share your feedback_id after running /bug and paste your settings.local.json file here - I can help validate. (please make sure that you are pre-autocompact with you run /bug)

It definitely fixed it for me. My json file was almost 400kb at the time. Removing it fixed the context left being at 4% all the time.

@Codewow do you still have your old settings.local.json - would love to peek at it

I understand @sid374 .. I need the MCPs to do my job.. I know 200K limit is beyond your control( hope you will raise to higher-ups). The back-to-back compaction is an issue... will capture /bug for you when it helps. Shall I do it after the 2nd compaction.

Start of sesssion:
 ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛀ ⛁ Context Usage
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ claude-opus-4-1-20250805 • 138k/200k tokens (69%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ System prompt: 17.2k tokens (8.6%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ System tools: 13.7k tokens (6.8%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ MCP tools: 95.9k tokens (47.9%)
⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ Custom agents: 1.9k tokens (1.0%)
⛀ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Memory files: 9.5k tokens (4.8%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ Messages: 8 tokens (0.0%)
⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ Free space: 61.9k (30.9%)

new funky error:

/compact

⎿ Error: Error during compaction: Error: API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.150:
tool_use ids were found without tool_result blocks immediately after: toolu_01McEXQALnVxhy2vYGxgh1L4. Each tool_use block must have
a corresponding tool_result block in the next message."},"request_id":"req_011CSVpaifmuxEeymwyBXYVM"}

/compact

⎿ Error: Error during compaction: Error: API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.150:
tool_use ids were found without tool_result blocks immediately after: toolu_01McEXQALnVxhy2vYGxgh1L4. Each tool_use block must have
a corresponding tool_result block in the next message."},"request_id":"req_011CSVpczmwtRKyXsDUkgpiR"}

╭────────────

new funky error: > /compact > ⎿ Error: Error during compaction: Error: API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.150: > tool_use ids were found without tool_result blocks immediately after: toolu_01McEXQALnVxhy2vYGxgh1L4. Each tool_use block must have > a corresponding tool_result block in the next message."},"request_id":"req_011CSVpaifmuxEeymwyBXYVM"} > /compact > ⎿ Error: Error during compaction: Error: API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.150: > tool_use ids were found without tool_result blocks immediately after: toolu_01McEXQALnVxhy2vYGxgh1L4. Each tool_use block must have > a corresponding tool_result block in the next message."},"request_id":"req_011CSVpczmwtRKyXsDUkgpiR"} ╭────────────

I do the same thing sometimes.
In this case I just execute /exit, and restart claude and continue on.

Ps
I have /autocompact disabled.
sent multiple /bug reports.

I tried to compact to to no effect:

/compact

⎿  Error: Compaction canceled.

· Debugging FAR-82 v15 workflow - no WhatsApp messages received… (esc to interrupt)
⎿  Next: Check latest execution logs for errors

should not error out:
["/Users/brtan/Projects/farmiere-api-integration/prod_workspaces/FAR-82/logToSupabase-NO-ZOKO-importable.json"])
⎿  {
"status": "calling_expert_analysis",
"step_number": 3,
… +148 lines (ctrl+r to expand)

⎿ API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"input length and max_tokens exceed context limit:
196348 + 21333 > 200000, decrease input length or max_tokens and try again"},"request_id":"req_011CSW3samzyy2BXVpNXPZh3"}

╭────

Unable to compact - create new issue?

/compact

⎿  Error: Error during compaction: Error: API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.149: all
messages must have non-empty content except for the optional final assistant message"},"request_id":"req_011CSYALERbhwoxVdQ2FBZ8n"}

/compact

⎿  Error: Error during compaction: Error: API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.149: all
messages must have non-empty content except for the optional final assistant message"},"request_id":"req_011CSYAMc9qUMMVMAW4vPfsM"}

⏺ Previous query still processing. Please try again.
⎿ API Error (529 {"type":"error","error":{"type":"overloaded_error","message":"Overloaded"},"request_id":null}) · Retrying in 33 seconds… (attempt 8/10)

@sid374 @benlrichards my god, I think they fix it...

<img width="1578" height="480" alt="Image" src="https://github.com/user-attachments/assets/249b50b3-3aa6-48d0-a518-63dee7e07c2e" />

Thanks @sid374 . MCP used to be at 50% . now it's 0% ... thanks for your help! 1.0.96

False alarm... had to re-login and same context % is there...

It kills me!! keep compacting coversation.. over and over!!! 1 ~2 lines and keep compacting!! disabled MCP, did multiple times /clear, started multiple times!!! it's near 0% and I can't keep using it. after 1,2 lines, keeps compacting! damn, tired!

Guys we having max plan for like ...4-5 months now . Compacting today 13 oct 2025 is insane!! 2-3 prompts , When reaching 8-10 % even a small question takes it to 0! . I would have no issue on this if compacting if the perfomance or if claude remembered properly the task.

We have noticed that you encountered the "Context auto-compact stuck at 0%" issue while using Claude Code, which is causing the tool to be unusable. According to your description, you have already tried clearing the cache, reinstalling Claude Code, and using the /clear command, but the issue remains unresolved. Additionally, you mentioned that multiple users have been affected by this problem.

To resolve this issue, we recommend that you try updating to the latest version of Claude Code. Please follow the steps below:

1. Open the terminal.
2. Run the command claude -v to confirm the current version.
3. If the current version is not the latest, please visit the official release page of Claude Code, download and install the latest version.
4. After installation is complete, restart Claude Code and test whether the issue is resolved.

If the problem still exists after updating to the latest version, please provide more detailed information so that we can assist you further. At the same time, we will also pay attention to feedback from other users and provide a solution as soon as possible.

Thank you for your patience and look forward to your reply.

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

Error compacting conversation

This issue has been automatically closed due to 60 days of inactivity. If you're still experiencing this issue, please open a new issue with updated information.

This issue was closed incorrectly despite recent human comments. This behavior of the bot is reported at https://github.com/anthropics/claude-code/issues/16497. Please upvote that issue, so maybe it gets noticed.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.