Crush keeps tool-result state after cancellation

[N0zoM1z0]

Description

Bug report

After SIGINT/cancellation is delivered during a streaming tool call, Crush still produces tool-result state for the interrupted response. A shell side-effect variant did not prove a real post-cancellation side effect, so this report is limited to post-cancellation tool-result handling.

Before/after behavior

Before: the provider stream crosses a duplicate, repeated, or cancelled tool-call boundary.

After: Crush records tool-result state that should have been rejected or ignored after cancellation.

Expected: After cancellation, Crush should drop interrupted tool-call state and should not return tool results for that response.

Minimal reproducible example

Standalone reproduction

Prerequisites: Docker, Python 3, and the GitHub CLI (gh) for the clone command below. The linked reproducer is self-contained and uses only Python standard-library modules plus Docker. It builds the affected CLI version from the public upstream package/release source and starts a local mock provider; it does not require this repository or any private fuzzing harness. The Docker run is limited to 2 CPUs and 4 GiB RAM by default.

Complete self-contained reproducer: https://gist.github.com/N0zoM1z0/453828f8978e9073d6fca24e21a57d91

The Gist contains crush-tool-executes-after-cancel.reproduce.py. Download and run:

gh gist clone 453828f8978e9073d6fca24e21a57d91 crush-tool-executes-after-cancel-reproducer
cd crush-tool-executes-after-cancel-reproducer
python3 crush-tool-executes-after-cancel.reproduce.py

To reuse an already-built local image:

python3 crush-tool-executes-after-cancel.reproduce.py --skip-build

Key output from a local run against the affected version:

process_exit=-9
provider_requests=69
tool_result_counts={'edit_cancel_1': 1756}
late_stream_events_after_stop=0
tool_result edit_cancel_1: observed=1756 expected>=1
REPRODUCED

Version

v0.76.0

Environment

• Repository: charmbracelet/crush - Version: Crush 0.76.0 - OS/arch: Linux x86_64 in Docker (node:24-bookworm) - Interface: CLI - Provider/model: OpenAI-compatible local mock provider, model gpt-4