fix(flows): inject artifact text, not the Part repr, into instructions - #7391

Please ensure you have read the contribution guide before creating a pull request.

Link to Issue or Description of Change

No existing issue; described below following the bug-report structure.

Describe the Bug:

{artifact.<name>} placeholders in instructions, and the Jinja artifact() helper, render str(artifact) (_instructions_utils.py, regex and Jinja paths). Every artifact service returns a types.Part, and str(Part) is its pydantic field dump. The existing tests use a mock load_artifact that returns a plain str, so they don't catch this.

Impact:

Every agent whose instruction references an artifact sends the model the Part repr instead of the content.

Observed Behavior:

With InMemoryArtifactService and types.Part(text="Buy milk."), the instruction Notes: {artifact.notes.txt} renders as:

Notes: media_resolution=None code_execution_result=None executable_code=None file_data=None function_call=None function_response=None inline_data=None text='Buy milk.' thought=None ...

Expected Behavior: Notes: Buy milk.

Solution:

Add _artifact_to_text and use it on both paths:

• a Part with text renders as that text;
• text/* inline_data is decoded as UTF-8;
• anything else keeps the existing str() rendering. Sending non-text artifacts as real parts is a separate feature (Images that are included into instructions via {artifact.<name>} syntax should allow a underlying LLM for image processing/interpeting. #4674).

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

Added test_inject_session_state_artifact_part_renders_as_text. It uses the real InMemoryArtifactService and is parametrized over the regex and Jinja paths and over text and text/plain inline-data parts. All 4 cases fail on main.

pytest tests/unittests/flows/llm_flows tests/unittests/utils   # 1505 passed

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end.
• Any dependent changes have been merged and published in downstream modules. (none)

🤖 Generated with Claude Code