Agent gets stuck in endless loop of "Model call failed"
Describe the bug
Unexpected failure loop shown in terminal output:
✓ write_bash
↪ 200 lines...
● Great! Now let's test if our new tool works:
✓ $ cd /home/rolfwr/mimir/alt && npx env-cmd esr server/tools/src/cmds/download-dynamodb-stream.ts /tmp/dynamodb-stream-test --max-hours=2 2>&1 | head -50
↪ 10 lines...
● The table exists but doesn't have a stream enabled according to the describe
response. Let me check the actual table structure:
✗ $ cd /home/rolfwr/mimir/alt && npx env-cmd aws dynamodb describe-table --table-name mb10-mimir-prod-genericTable 2>&1 | grep -i stream | head -10
Operation aborted by user
> A point of clarification. The table we care about, which has streams enabled is "mb10-mimir-prod-item".
✗ Model call failed: {"message":"messages.192: `tool_use` ids were found without `tool_result` blocks immediately after: toolu_01UvmZN5EkHfndM9X2qekRsA. Each `tool_use` block must have a
corresponding `tool_result` block in the next message.","code":"invalid_request_body"} (Request ID: 85D4:239294:154F760:195108A:68F23588)
✗ Model call failed: {"message":"messages.192: `tool_use` ids were found without `tool_result` blocks immediately after: toolu_01UvmZN5EkHfndM9X2qekRsA. Each `tool_use` block must have a
corresponding `tool_result` block in the next message.","code":"invalid_request_body"} (Request ID: 85D4:239294:1550F49:1952BF3:68F2358A)
✗ Model call failed: {"message":"messages.192: `tool_use` ids were found without `tool_result` blocks immediately after: toolu_01UvmZN5EkHfndM9X2qekRsA. Each `tool_use` block must have a
corresponding `tool_result` block in the next message.","code":"invalid_request_body"} (Request ID: 85D4:239294:15529EB:1954AF1:68F23591)
✗ Model call failed: {"message":"messages.192: `tool_use` ids were found without `tool_result` blocks immediately after: toolu_01UvmZN5EkHfndM9X2qekRsA. Each `tool_use` block must have a
corresponding `tool_result` block in the next message.","code":"invalid_request_body"} (Request ID: 85D4:239294:15544FC:1956AB8:68F2359A)
✗ Model call failed: {"message":"messages.192: `tool_use` ids were found without `tool_result` blocks immediately after: toolu_01UvmZN5EkHfndM9X2qekRsA. Each `tool_use` block must have a
corresponding `tool_result` block in the next message.","code":"invalid_request_body"} (Request ID: 85D4:239294:1556D5E:19598B7:68F235A2)
✗ Model call failed: {"message":"messages.192: `tool_use` ids were found without `tool_result` blocks immediately after: toolu_01UvmZN5EkHfndM9X2qekRsA. Each `tool_use` block must have a
corresponding `tool_result` block in the next message.","code":"invalid_request_body"} (Request ID: 85D4:239294:155C27F:195FB5D:68F235AD)
> You seem to have gotten stuck.
✗ Model call failed: {"message":"messages.192: `tool_use` ids were found without `tool_result` blocks immediately after: toolu_01UvmZN5EkHfndM9X2qekRsA. Each `tool_use` block must have a
corresponding `tool_result` block in the next message.","code":"invalid_request_body"} (Request ID: 985E:107A64:12E48E5:162D645:68F23643)
✗ Model call failed: {"message":"messages.192: `tool_use` ids were found without `tool_result` blocks immediately after: toolu_01UvmZN5EkHfndM9X2qekRsA. Each `tool_use` block must have a
corresponding `tool_result` block in the next message.","code":"invalid_request_body"} (Request ID: 985E:107A64:12E5D57:162EE4A:68F23644)
Problem persists after exiting out of copilot and re-running copilot with --continue option.
Affected version
0.0.343 Commit: 5847051
Steps to reproduce the behavior
Reproducing it using "copilot --continue" would require a way of exporting whatever persistent state that "--continue" operates on.
Expected behavior
Copilot does not go into endless loop of reporting ✗ Model call failed: {"message":"messages.192: tool_useids were found withouttool_resultblocks immediately after: toolu_01UvmZN5EkHfndM9X2qekRsA. Eachtool_useblock must have a correspondingtool_result block in the next message.","code":"invalid_request_body"} (Request ID: 847A:20242C:13286FE:1670E7E:68F2366E)
Or if it gets into this state, there is a way to reset it and get it back on track.
Additional context
No response
Experiencing this a lot too after recently updating - it seems to happen when you abort an in progress command. So to reproduce I think you just need to get it to run a script that takes a while and then abort it. Type your comment. It then fails.
I am getting same. To reproduce bug
- Give some prompt
- When tool call such as search or read is used, stop it from further generating
- Add prompts further below that (stopping is intentional to refine prompt directing copilot to look correct place)
As in screenshot, just when tool ended, I pressed stop. When I typed further prompts, got infinite error loops.
Suggested solution: Mark tool call as failed when stop button is pressed, in chat history database.