Add Langfuse tracing to ICD Agent and export traces as DSPy training/evaluation inputs

[filiperochalopes]

User story

As a developer, I want improved observability for the ICD Agent by capturing full execution traces so we can reuse real runs as DSPy datasets for evaluation and optimization.

Use case

Run the ICD Agent in normal operation, store all traces in Langfuse, then export those traces to a structured dataset to feed DSPy.

Requirements

• Integrate Langfuse tracing into the ICD Agent execution flow.
• Capture and store all relevant traces for each run (end-to-end) in Langfuse.
• Provide an export process to convert stored traces into DSPy-compatible input data.

Acceptance criteria

• Given an ICD Agent execution, a complete trace is recorded in Langfuse for that run.
• Traces can be exported from Langfuse into a dataset format usable as input for DSPy.
• Exported dataset contains enough information to reproduce/score runs in DSPy (based on the captured traces).

[ocl-automation]

Automated Triage — 2026-05-15

Dimension	Score
Complexity	4
Scope	4
Clarity	3
Dependency	2
Risk	3

Decision: decision/planning-needed
Type: type/feature
Signals: signal/large-scope

Complex integration requiring design decisions for trace capture points and export format across multiple ICD Agent components. Moderate clarity with clear goal but missing implementation details like trace structure and DSPy format requirements. Moderate risk as incorrect tracing could impact agent performance or data quality.

---

Generated by ticket-triage pipeline v3. Model: anthropic/claude-sonnet-4 (openrouter).