Adding image mid-turn triggers Context limit reached + compaction failure

Resolved 💬 3 comments Opened Feb 16, 2026 by webexpr-jdevesvrotte Closed Feb 16, 2026

Bug Description

Adding an image to the conversation while Claude is actively processing (mid-turn) immediately triggers "Context limit reached · /compact or /clear to continue". Then attempting /compact fails with:

Error: Error during compaction: Error: Conversation too long. Press esc twice to go up a few messages and try again.

The only recovery option is /clear, which loses the entire conversation context.

Steps to Reproduce

  1. Start a conversation with Claude Code
  2. Have Claude perform a multi-step task (e.g., browser automation with MCP tools)
  3. While Claude is actively processing (spinner visible, tool calls in progress), paste/add an image
  4. Observe "Context limit reached" error
  5. Try /compact → fails with "Conversation too long"
  6. Only /clear works

Expected Behavior

  • Adding images mid-turn should either queue the image for the next turn or handle it gracefully
  • At minimum, /compact should still work to recover the conversation
  • Previously, adding images mid-turn worked without issues

Actual Behavior

  • Immediate "Context limit reached" error
  • /compact fails completely
  • Conversation is unrecoverable without /clear

Important Context

  • Adding images when Claude is not processing (idle/waiting for input) works perfectly fine, even multiple images
  • This is a recent regression — previously it was possible to add images mid-turn without issues

Screenshots

Context limit reached after adding image mid-turn:

The image was added right after a navigate_page MCP tool response, while Claude was still processing.

Compaction failure:

/compact returns: Error: Error during compaction: Error: Conversation too long. Press esc twice to go up a few messages and try again.

Environment

  • Claude Code CLI (latest)
  • macOS
  • Using MCP servers (chrome-devtools)

View original on GitHub ↗

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