Measure agent concurrency limits

measure concurrency cap via stress probe

empirically validating the workflow runtime's true parallel capacity before launching production workloads

• · N (integer, default 30)

• · { N, ok, sample: results[0:6] }

• · parallel() orchestrator
• · agent() executor
• · haiku model

• · workflow runtime with parallel() + agent() + log()
• · haiku model available

• · agent null returns reduce ok count
• · concurrency cap differs from expected min(16,cores-2)

• · hardcoded expected cap formula min(16,cores-2)
• · trivial agents minimize cost while exercising real scheduler