ci: forward runner arguments

[joshblack]

Follow up to #203.

This PR applies the runner feedback that arrived as #203 was merging and is updated for the durable plan sharding added in #204. Benchmark and experiment names remain required positional arguments, callers can optionally select run, plan, shard, or merge, and any remaining arguments are forwarded to the underlying agent-eval command.

Options can be passed directly after the name for the default run mode or after an explicit mode. As a result, every runner mode supports options such as --log-level without editing the wrapper scripts.

This is a change to our internal workflow and has no public facing impact.

Changelog

New

• None.

Changed

• Update the benchmark and experiment runners to accept additional agent-eval options in every execution mode.
• Preserve the run, plan, shard, and merge behavior now available on main.
• Update runner usage messages to document modes and optional CLI arguments.

Removed

• Remove the argument-count restriction that rejected additional runner options.

[Copilot]

Copilot review overview

🟡 Changes recommended

The new mode-detection logic misclassifies short agent-eval flags (e.g., -h, -c) as modes, breaking the stated goal of forwarding options in every mode.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Review tier: Lite
Findings: 2

New issues introduced by this change (2)

Severity	Finding
	script/​run-benchmark.sh — Mode detection treats any first non---* token as a mode, so passing short agent-eval options…
	script/​run-experiment.sh — Mode detection treats any first non---* token as a mode, so passing short agent-eval options…

What changed in this PR

This PR updates the internal benchmark and experiment runner shell scripts to accept and forward additional agent-eval CLI options across all runner modes, while keeping the existing run|plan|shard|merge workflows introduced in earlier stacked PRs.

Changes:

• Relax runner argument-count checks and update usage text to document forwarding of additional agent-eval options.
• Add mode parsing that supports an optional explicit mode while defaulting to run.
• Forward remaining CLI arguments ("$@") into each underlying agent-eval invocation.

File	Description
script/​run-benchmark.sh	Adds mode parsing and forwards remaining args to agent-eval for benchmark runs across all modes.
script/​run-experiment.sh	Adds mode parsing and forwards remaining args to agent-eval for experiment runs across all modes.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.