Crush sends duplicate tool-result history for a duplicate tool-call id

[N0zoM1z0]

Description

Bug report

When the provider emits duplicate tool-call id dup_id_0001, Crush appends duplicate role=tool results for that id into later OpenAI Chat request history. This can corrupt provider conversation state, trigger provider-side schema errors, or amplify retries.

Before/after behavior

Before: the local mock provider emits two tool calls with the same id, dup_id_0001.

After: Crush sends duplicate tool-result history for dup_id_0001 in a subsequent provider request. In the latest local verification run, the reproducer observed 2228 duplicate entries for that id before timeout.

Expected: Crush should reject, ignore, or coalesce duplicate tool-call ids before executing tools or appending results to provider history.

Minimal reproducible example

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 @charmland/crush@0.76.0 from the public npm package, starts a local mock provider, and runs the CLI in an isolated workspace. The Docker run is limited to 2 CPUs and 4 GiB RAM by default.

Complete self-contained reproducer: https://gist.github.com/N0zoM1z0/1ac3426608603aa4cd3249a40331614e

gh gist clone 1ac3426608603aa4cd3249a40331614e crush-duplicate-tool-id-history-pollution-reproducer
cd crush-duplicate-tool-id-history-pollution-reproducer
python3 crush-duplicate-tool-id-history-pollution.reproduce.py

To reuse an already-built local image:

python3 crush-duplicate-tool-id-history-pollution.reproduce.py --skip-build

Expected successful reproduction output includes:

provider_requests=1125
duplicate_tool_result_in_request dup_id_0001: observed=2228 expected>=2
REPRODUCED

Version

0.76.0

Environment

OS/arch: Linux x86_64 Docker, using node:24-bookwormbased target image ; Interface: CLI;Provider/model: OpenAI Chat-compatible local mock provider, gpt-4