v0.1.9

What's Changed

• fix: deep-merge dict metadata during aggregation by @declanjackson in #41
• fix(clients): avoid empty text parts when encoding OpenAI message content by @declanjackson in #50
• feat: allow passthrough kwargs to E2B sandbox creation by @declanjackson in #42
• feat: add create_gate kwarg to E2BCodeExecToolProvider for throttling sandbox creation by @BillyDodds in #48
• feat: allow multiple finish tools by @declanjackson in #49
• fix: serialize Docker image prepare to avoid concurrent-pull race by @BillyDodds in #47
• Fix/cache handling by @MohamedAkbarally in #36
• fix: prevent zombie process leak on code_exec timeout by @BillyDodds in #4

⚠️ Breaking changes

full_msg_history no longer appends per turn

session.run() returns full_msg_history: list[list[ChatMessage]]. As of this release, a new group is only appended on context summarization (plus one final
group at session end) — not once per turn. len(full_msg_history) no longer reflects the agent's turn count.

To count turns, count AssistantMessage instances across the flattened history:

from itertools import chain
from stirrup.core.models import AssistantMessage

num_turns = sum(
    1 for m in chain.from_iterable(full_msg_history) if isinstance(m, AssistantMessage)
)

New Contributors

• @BillyDodds made their first contribution in #48

Full Changelog: v0.1.8...v0.1.9