[lsmlhi_25 (FOWEPJF255)]

Summary

• Fixes #8955: streamed code_interpreter input/output deltas were silently dropped when one chunk's text was a string prefix of another (e.g. "(" then "()", or indent whitespace then the indented line).
• Root cause: _merge_content_item_lists treated any prefix relationship as an OpenAI-style full resend and discarded one side.
• Fix: always concatenate ordinary deltas; only replace accumulated text when the incoming payload is an authoritative complete value (*.done raw_representation.type, or stream_complete / complete on additional_properties).
• Same heuristic also caused #8903; this change covers that path as well.

Test plan

• python -m pytest tests/core/test_types.py -k "code_interpreter_tool_call_stream or code_interpreter_tool_call_content or code_interpreter_tool_result_content" -q
• CI

Closes #8955

[Copilot]

Copilot review overview

🟡 Changes recommended

Empty lists can raise IndexError, and an existing session-history test now fails.

Review effort: Balanced
Findings: 2

Open (2)

• Empty accumulated list causes IndexError · New
• Unconditional concatenation breaks session-history reconstruction · New

What changed in this PR

Fixes corruption when aggregating prefix-overlapping code-interpreter stream deltas.

Changes:

• Concatenates ordinary deltas unconditionally.
• Replaces accumulated text for explicitly completed payloads.
• Adds regression tests for prefix overlap and done events.

File	Description
python/​packages/​core/​agent_framework/​_types.py	Updates code-interpreter stream merging.
python/​packages/​core/​tests/​core/​test_types.py	Adds aggregation regression tests.

---

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

[lsmlhi_25 (FOWEPJF255)]

Thanks Copilot -- addressed both findings in 772ad82f:

1. Empty accumulation -- if existing is an empty list (e.g. initial outputs=[]), return a copy of incoming before indexing [0]. Added test_code_interpreter_tool_result_empty_outputs_then_text.
2. Session-history fixture -- marked the final full-code snapshot with stream_complete=True so it replaces deltas instead of concatenating. test_after_run_stores_coalesced_code_interpreter_chunks passes again.

[lsmlhi_25 (FOWEPJF255)]

Thanks Eduard van Valkenburg (@eavanvalkenburg) -- agreed. In 463ae934:

• Core _content_items_are_complete_value only looks at the generic additional_properties["stream_complete"] / "complete" flags (no raw_representation / *.done inspection).
• OpenAI's code_interpreter_call_code.done path now sets stream_complete=True on the emitted content so the replace-vs-concat behavior stays correct at the provider boundary.