Add OpenCode CLI-backed agent runner for AssetOpsBench

[ChathurangiShyalika]

Summary

This PR adds an opencode-agent runner so AssetOpsBench can evaluate OpenCode as a CLI-based agent alongside the existing SDK-backed agents.

What Changed

• Added a new agent.opencode_agent package with:

  • OpenCodeAgentRunner
  • opencode-agent CLI entry point
  • OpenCode runtime config generation
  • OpenCode JSON event parsing into AssetOpsBench trajectories
  • Trajectory persistence through the existing AGENT_TRAJECTORY_DIR flow

• Registered the new CLI command:

opencode-agent = "agent.opencode_agent.cli:main"

• Added opencode_agent as a supported benchmark method in scenario_suite_runner.

• Configured OpenCode MCP access to the existing AssetOpsBench MCP servers

Usage

Run OpenCode directly:

uv run opencode-agent \
  --model-id tokenrouter/MiniMax-M3 \
  --show-trajectory \
  "What is the current time?"

Run through the benchmark suite:

uv run python -m benchmark.scenario_suite_runner \
  --scenario-ids benchmarks/scenario_suite/scenarios.txt \
  --scenario-root /path/to/scenarios_data \
  --agent_name opencode_agent \
  --model-id tokenrouter/MiniMax-M3

Testing

Unit tests:

.venv/bin/pytest src/agent/opencode_agent/tests/test_runner.py

Result:

11 passed

CLI smoke test:

.venv/bin/opencode-agent --help

Manual benchmark check:

Ran scenario 401 with tokenrouter/MiniMax-M3

OpenCode produced a trajectory and report
Final answer matched the ground truth: C