fix: escalation memory spans render wrong type and empty results

[mjnovice]

Problem

Customer feedback (Schneider, via #dev-agent-memory): the escalation memory span in the execution trace shows type "Dynamic few-shot" and nothing in the Results tab.

Root cause, verified against @uipath/traceview (2.45.0 and current traceview-ui main):

1. The escalation lookup emitted an "Apply escalation memory" child span typed applyDynamicFewShot, borrowed from the episodic memory flow. The UI labels that type "Dynamic few-shot".
2. The UI renderer for applyDynamicFewShot only displays episodic feedback items — it requires response.results[].feedback.comment. Escalation search results are shaped results[].answer.{output, outcome} (no feedback field ever), so the Results tab could never show anything.
3. Latent bug: the "Save escalation memory" ingest span used kind agentMemoryWrite, which is not in the UI span-type registry and renders as type "N/A". The registered type is agentMemoryStore.

The LlmOps backend does not key any processing off these span kinds — this is purely a UI rendering contract.

Changes

• Lookup: drop the applyDynamicFewShot child span and collapse onto the single "Find previous memories" span (agentMemoryLookup → renders as "Memory lookup"), now carrying request, response, fromMemory, memoryItemsMatched, and result. On a cache hit the cached answer also remains visible as the escalation tool span's own output.
• Ingest: retype agentMemoryWrite → agentMemoryStore ("Memory store"), and emit inputs/outputs attributes — exactly what the UI's default renderer displays in the Results tab — so the save span shows the persisted memory item and savedToMemory outcome.

Notes

With today's traceview there is no span type that renders escalation-memory matches in a dedicated Results tab (agentMemoryLookup is deliberately excluded from that tab, same as episodic lookups). Rich rendering of the matched item needs a small traceview-ui follow-up (e.g. a renderer for results[].answer or a new applyEscalationMemory type). This PR fixes the wrong label and the misleading empty tab within the current contract.

Testing

• uv run pytest tests/agent/ — 1273 passed
• uv run ruff check / ruff format --check / mypy on touched files — clean

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR fixes how escalation-memory tracing spans are typed and populated so the Studio/traceview UI renders the correct span type labels and shows meaningful Results content for ingest operations. It aligns emitted span kinds and attributes with the current UI rendering contract, without changing backend behavior.

Changes:

• Remove the applyDynamicFewShot child span from escalation lookup and consolidate lookup details onto the single agentMemoryLookup (“Memory lookup”) span.
• Retype escalation ingest span from agentMemoryWrite to agentMemoryStore and emit inputs/outputs attributes so the UI Results tab can display persisted content and outcome.
• Update escalation-memory tests to assert the new span structure, kinds, and attributes.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
tests/agent/tools/test_escalation_memory.py	Updates span assertions to match the new lookup span shape and the ingest span kind/attributes (agentMemoryStore, inputs/outputs).
src/uipath_langchain/agent/tools/escalation_memory.py	Adjusts tracing spans for escalation memory lookup/ingest to match UI span-type registry and renderer expectations.

[saksharthakkar]

Do you need to bump up langchain version?

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud