RFC: a fault-injection reliability benchmark for agent runs

#488 and the recurring sandbox-restart failures (#14260, #13578) are all symptoms of the same gap: we have no systematic way to measure whether an agent run survives faults. The SDK paper frames divergent agent/sandbox state and mid-tool-call crashes as the core V1 reliability problem, but there's no benchmark that actually stresses it.

I'd like to build one. The idea: a harness that runs real agent tasks (a SWE-bench subset to start) while injecting a defined fault schedule — sandbox restart, lost response, SIGKILL mid-tool-call, network partition — and scores three things: does the run resume and complete, does it avoid re-executing an already-committed irreversible action on replay, and how much overhead recovery costs. Output is a reliability scorecard per agent/config.

Context on me: I recently open-sourced a reconciliation engine for irreversible agent actions (persist-before-dispatch, reconcile-by-query, preregistered fault-schedule benchmark), so this is squarely what I've been building. Before I invest in it, is this something the team wants, and does it belong in benchmarks or closer to the SDK? Happy to scope it down to a first milestone if that's easier to land.

Start by reviewing #488 and the recurring sandbox-restart failures in #14260 and #13578, then compare the proposed SWE-bench subset and fault schedule with the existing benchmarks harness. The issue does not name files, tests, or an agreed first milestone; completion criteria remain to be defined around recovery, replay safety, and overhead scoring.