Skip to content

[Bug]: panic: runtime error: index out of range [-1] #540

@VikasChoataHP

Description

@VikasChoataHP

Environment (please complete the following):

  • OS: Darwin Kernel Version 24.6.0
  • kubectl-ai version:
    - version: 0.0.23
    - commit: 0f6537e
    - date: 2025-08-25T23:13:54Z
  • LLM provider: openai
  • LLM model: claude sonnet 4

Describe the bug

I get intermittent error using kubectl-ai on mac machine.

  panic: runtime error: index out of range [-1]
  
  goroutine 8 [running]:
  github.com/openai/openai-go.(*ChatCompletion).accumulateDelta(_, {{0x140004741c7, 0x11}, {0x14000634288, 0x1, 0x1}, 0x68c978f5, {0x140004741f8, 0x63}, {0x14000474349, ...}, ...})
  	/home/runner/go/pkg/mod/github.com/openai/openai-go@v1.11.0/streamaccumulator.go:129 +0x6e0
  github.com/openai/openai-go.(*ChatCompletionAccumulator).AddChunk(_, {{0x140004741c7, 0x11}, {0x14000634288, 0x1, 0x1}, 0x68c978f5, {0x140004741f8, 0x63}, {0x14000474349, ...}, ...})
  	/home/runner/go/pkg/mod/github.com/openai/openai-go@v1.11.0/streamaccumulator.go:40 +0x4c
  github.com/GoogleCloudPlatform/kubectl-ai/gollm.(*openAIChatSession).SendStreaming.func1(0x140004e0640)
  	/home/runner/work/kubectl-ai/kubectl-ai/gollm/openai.go:332 +0x128
  github.com/GoogleCloudPlatform/kubectl-ai/pkg/agent.(*Agent).Run.func1()
  	/home/runner/work/kubectl-ai/kubectl-ai/pkg/agent/conversation.go:498 +0x10a0
  created by github.com/GoogleCloudPlatform/kubectl-ai/pkg/agent.(*Agent).Run in goroutine 1
  	/home/runner/work/kubectl-ai/kubectl-ai/pkg/agent/conversation.go:293 +0x1b0

I did ask LLM about it, here is what I got back

"“When using streaming responses from the OpenAI-compatible model, an occasional malformed incremental update (a chunk with an invalid internal index) causes the underlying OpenAI Go SDK to panic with ‘index out of range [-1]’. This is not caused by your input; it’s a transient protocol issue. Re-running the command usually succeeds because the next streamed sequence is well-formed.”"

Metadata

Metadata

Assignees

Labels

bugSomething isn't working

Type

No type
No fields configured for issues without a type.

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions