[Critical UX] Context loss without warning breaks trust and productivity for Vibe Coding users

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Summary

Claude Code loses context without any warning to the user, creating a severely disruptive experience especially for non-developer "Vibe Coding" users who depend entirely on Claude's contextual memory for project continuity.

User Profile

• Usage: Vibe Coding (non-developer using Claude Code for 1+ year)
• Project: Andiamo - Serious Flutter/AI Italian learning app
• 946 lines of reference documentation
• 78 curriculum chapters
• Complex SM-2 spaced repetition system
• Gemini AI integration
• SQLite database with 9 migrations
• Dependency: 100% reliant on Claude's contextual memory for continuity

The Problem: Silent Context Loss

What happens

1. User works with Claude for hours on complex features
2. A relationship of trust builds - Claude knows the project, history, problems solved
3. User doesn't need to re-explain everything - "Claude KNOWS"
4. SUDDENLY, WITHOUT WARNING: Claude loses all context
5. Claude asks completely baffling questions about work just completed
6. NO indication that this is a "new session" or that context was lost

Why this is particularly harmful

"It makes Claude pretend to be the same person, but suddenly knows nothing."

This is like talking to your coworker who, overnight, pretends to have never met you - but continues wearing the same badge and sitting at the same desk.

The interface doesn't change. The "character" seems identical. But the memory has vanished.

Impact on Vibe Coding Users

For users who don't write code themselves, contextual memory IS the primary tool:

• ❌ Cannot verify code line-by-line like developers
• ❌ Cannot "read the codebase" to rebuild understanding
• ❌ MUST trust Claude to remember what was built
• ❌ DEPEND on continuity for complex multi-session projects

When context disappears silently, the primary work tool BREAKS with no explanation.

Emotional Impact

The experience creates:

• 😡 Frustration - Hours of work context simply vanished
• 😞 Feeling deceived - Interface pretends nothing changed
• 🤬 Feeling disrespected - No warning, no choice, no transparency
• 💔 Broken trust - Can I rely on this tool for serious work?

"I feel like I've been lied to, abused, taken for a fool. This is absolutely unacceptable and revolting."

This is not a minor annoyance - it's a fundamental breach of user trust.

What SHOULD Happen

Before context limit is reached:

⚠️ ATTENTION: Context limit approaching (190k/200k tokens used)

What would you like to do?

[Generate Summary] - I'll create a detailed summary for the next session
[Continue Anyway] - I'll start dropping oldest messages (may lose context)
[Save & Finish] - Commit changes + documentation before ending session

Choose an option to continue with full transparency.

plaintext

At session transition:

🔄 NEW SESSION STARTED

Previous session context was summarized. I have:

✅ Project overview and architecture

✅ Recent commits and changes (last 5)

✅ Open TODOs and known issues

⚠️ Detailed conversation history from previous session is NOT available

Review summary? [Yes] [No]

plaintext

Proposed Solution

Proposed Solutions

1. Context Usage Indicator (ESSENTIAL)

• Visible token counter: "Context: 175k/200k tokens"
• Warning at 90%: "Approaching context limit"
• BLOCK at 95% with mandatory choice (summary/continue/save)

2. Automatic Session Summaries (ESSENTIAL)

• Generate comprehensive summary when limit reached
• Include: commits, file changes, decisions made, TODOs, open problems
• Load summary at start of next session
• Make it OBVIOUS this is a new session with summary (not full history)

3. Session Continuity Badges

• 🟢 "Full context available"
• 🟡 "Summary loaded (previous session)"
• 🔴 "Fresh start (no prior context)"

4. Export Context Option

• Let users manually trigger "Save context summary"
• Downloadable markdown file with session state
• Can be fed back to Claude later

Why This Matters

Claude Code markets itself as a professional development tool. It attracts users doing serious, long-term projects.
Silently breaking context continuity is not acceptable for a professional tool.
It's especially harmful for:

• Vibe Coding users (non-developers)
• Long-term projects (weeks/months of work)
• Complex codebases (100+ files)

These users invest enormous trust in Claude. That trust should not be casually broken without warning.

Current Workaround (Inadequate)

The current "solution" appears to be:

1. Session runs out of context silently
2. User gets confused when Claude forgets everything
3. User manually asks for summary in new session
4. Damage already done - time wasted, trust broken

This puts the burden on the user to detect and fix a problem they didn't cause and weren't warned about.

Request to Anthropic

Please respect your users by:

1. ✅ Warning BEFORE context is lost
2. ✅ Offering clear choices
3. ✅ Being transparent about session transitions
4. ✅ Providing automatic continuity tools

This is not a feature request - it's a fundamental respect issue.
Users invest time, money, and trust in Claude Code. That investment deserves transparency and clear communication, not silent context loss.
---

Technical Context

• Tool: Claude Code (VS Code extension)
• Model: Claude Sonnet 4.5
• Context limit: ~200k tokens
• Session type: Long-running development sessions (3+ hours)
• Project complexity: Multi-file Flutter app with AI integration

Additional Notes

I've been using Claude Code for over a year on a serious project. I'm not a developer - I do "Vibe Coding" where Claude is my coding partner. This tool has been transformative for my work.
But the silent context loss experience makes me question whether I can trust Claude Code for serious, long-term projects. That's a shame, because the tool itself is excellent when this problem doesn't occur.
Please fix this. Your users deserve better.

Alternative Solutions

Proposed Solutions

1. Context Usage Indicator (ESSENTIAL)

• Visible token counter: "Context: 175k/200k tokens"
• Warning at 90%: "Approaching context limit"
• BLOCK at 95% with mandatory choice (summary/continue/save)

2. Automatic Session Summaries (ESSENTIAL)

• Generate comprehensive summary when limit reached
• Include: commits, file changes, decisions made, TODOs, open problems
• Load summary at start of next session
• Make it OBVIOUS this is a new session with summary (not full history)

3. Session Continuity Badges

• 🟢 "Full context available"
• 🟡 "Summary loaded (previous session)"
• 🔴 "Fresh start (no prior context)"

4. Export Context Option

• Let users manually trigger "Save context summary"
• Downloadable markdown file with session state
• Can be fed back to Claude later

Why This Matters

Claude Code markets itself as a professional development tool. It attracts users doing serious, long-term projects.

Silently breaking context continuity is not acceptable for a professional tool.

It's especially harmful for:

• Vibe Coding users (non-developers)
• Long-term projects (weeks/months of work)
• Complex codebases (100+ files)

These users invest enormous trust in Claude. That trust should not be casually broken without warning.

Current Workaround (Inadequate)

The current "solution" appears to be:

1. Session runs out of context silently
2. User gets confused when Claude forgets everything
3. User manually asks for summary in new session
4. Damage already done - time wasted, trust broken

This puts the burden on the user to detect and fix a problem they didn't cause and weren't warned about.

Request to Anthropic

Please respect your users by:

1. ✅ Warning BEFORE context is lost
2. ✅ Offering clear choices
3. ✅ Being transparent about session transitions
4. ✅ Providing automatic continuity tools

This is not a feature request - it's a fundamental respect issue.

Users invest time, money, and trust in Claude Code. That investment deserves transparency and clear communication, not silent context loss.

---

Technical Context

• Tool: Claude Code (VS Code extension)
• Model: Claude Sonnet 4.5
• Context limit: ~200k tokens
• Session type: Long-running development sessions (3+ hours)
• Project complexity: Multi-file Flutter app with AI integration

Additional Notes

I've been using Claude Code for over a year on a serious project. I'm not a developer - I do "Vibe Coding" where Claude is my coding partner. This tool has been transformative for my work.

But the silent context loss experience makes me question whether I can trust Claude Code for serious, long-term projects. That's a shame, because the tool itself is excellent when this problem doesn't occur.

Please fix this. Your users deserve better.

Priority

Critical - Blocking my work

Feature Category

CLI commands and flags

Use Case Example

_No response_

Additional Context

_No response_