fix(session): prevent assistant-last message array from reaching provider APIs

[brendandebeasi]

Summary

• Fixes "This model does not support assistant message prefill" errors that occur when the model message array ends with an assistant message
• Adds a guard in toModelMessages that appends a synthetic user message when the filtered array ends with an assistant role

Problem

After compaction, error recovery, or when filtered-out user messages leave an assistant message at the tail of the model message array, some providers and the AI SDK reject the conversation. This manifests as:

AI_APICallError: This model does not support assistant message prefill. 
The conversation must end with a user message.

Observed with anthropic/claude-opus-4-6 where the error comes from the AI SDK's message validation before reaching the Anthropic API.

Solution

Added a guard at the end of toModelMessages() in message-v2.ts that detects when the filtered message array ends with an assistant role and appends a synthetic "Continue." user message to satisfy provider requirements.

This complements the existing synthetic user message injection in prompt.ts (line 506) which handles the task/command case but doesn't cover all edge cases (compaction, error recovery, etc.).

Related

• Existing workaround in prompt.ts:505-528 for task command case
• Affects all providers that don't support assistant prefill

[github-actions]

This PR doesn't fully meet our contributing guidelines and PR template.

What needs to be fixed:

• PR description is missing required template sections. Please use the PR template.

Please edit this PR description to address the above within 2 hours, or it will be automatically closed.

If you believe this was flagged incorrectly, please let a maintainer know.

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Based on the search results, I found two potentially related PRs:

Potential Duplicates/Related PRs:

1. PR fix: disable assistant prefill for Claude 4.6 models #14772 - "fix: disable assistant prefill for Claude 4.6 models"

   • Related to the same issue of Claude models rejecting assistant message prefill
   • May address part of the same problem for a specific provider

2. PR fix: keep Copilot Claude prompts user-final across continuation turns #16921 - "fix: keep Copilot Claude prompts user-final across continuation turns"

   • Related to ensuring conversations end with user messages (user-final)
   • Addresses a similar pattern of managing message endings across turns

3. PR fix(session): prevent infinite loop in auto-compaction when assistant ended its turn #15532 - "fix(session): prevent infinite loop in auto-compaction when assistant ended its turn"

   • Related to the compaction edge case mentioned in your PR description as contributing to the issue

The current PR (18421) appears to be a more comprehensive solution addressing the root cause across all providers, whereas the earlier PRs may have been targeted fixes for specific scenarios.

[github-actions]

This pull request has been automatically closed because it was not updated to meet our contributing guidelines within the 2-hour window.

Feel free to open a new pull request that follows our guidelines.