feat(instrument): typed events Bundle #3 — google_adk + llama_index + ms_agent_framework by mmercuri · Pull Request #151 · LayerLens/stratix-python

mmercuri · 2026-05-10T18:01:42Z

Summary

Migrates the next 3 framework adapters from emit_dict_event(...) to typed
emit_event(TypedModel.create(...)) against the canonical Pydantic models
from layerlens.instrument._compat.events (PR #129 foundation).

This is Bundle #3 of 6 in the typed-events migration series. Bundle
#1 (PR #129) shipped the foundation + agno reference. Bundle #2 (PR #138)
shipped autogen + crewai + smolagents. This bundle ships google_adk +
llama_index + ms_agent_framework.

Honest site counts (grep, not estimated)

Per CLAUDE.md item 11 — counts verified with
grep -E 'emit_dict_event' src/.../<adapter>/:

Adapter	Migration doc estimate	Actual grep count
`google_adk`	~11	11
`llama_index`	~12	12
`ms_agent_framework`	~12	12
Total	~35	35

Estimates were honest in this bundle.

Per-adapter status

google_adk (commit `ffa1aca`)

11 emission sites migrated
Helpers added: _stringify, _sha256_of, _coerce_to_dict,
_detect_provider
mypy --strict: clean (2 source files)
pytest: 15/15 passing
Regression tests added: typed-payloads-only assertion +
no-DeprecationWarning assertion

llama_index (commit `4fc3a29`)

12 emission sites migrated
Retrieval events mapped onto ToolCallEvent with name="retrieval",
integration=IntegrationType.LIBRARY (no canonical retrieval shape;
mirrors agno reference)
Helpers added: _stringify, _coerce_to_dict, _sha256_of,
_detect_provider
mypy --strict: clean (2 source files)
pytest: 15/15 passing
Regression tests added

ms_agent_framework (commit `0917481`)

12 emission sites migrated
Behavioural change: the previous adapter emitted an ad-hoc
agent.state.change payload alongside agent.output carrying a
run_complete / run_failed marker. That payload did not satisfy
the canonical AgentStateChangeEvent before_hash / after_hash
contract (no real state mutation to hash at the run boundary).
The post-migration mapping carries the same signal as run_status
on AgentOutputEvent.metadata — preserves the cross-cutting
completion marker without violating the canonical schema.
Helpers added: _stringify, _coerce_to_dict, _sha256_of
mypy --strict: clean (2 source files)
pytest: 16/16 passing
Regression tests added

Combined verification

$ uv run mypy --strict src/layerlens/instrument/adapters/frameworks/google_adk \
    src/layerlens/instrument/adapters/frameworks/llama_index \
    src/layerlens/instrument/adapters/frameworks/ms_agent_framework
Success: no issues found in 6 source files

$ uv run python -m pytest \
    tests/instrument/adapters/frameworks/test_google_adk_adapter.py \
    tests/instrument/adapters/frameworks/test_llama_index_adapter.py \
    tests/instrument/adapters/frameworks/test_ms_agent_framework_adapter.py
46 passed in 1.05s

Multi-tenancy compliance

Per PR #129 foundation: BaseAdapter.emit_event stamps org_id onto
every typed payload before delegating to self._stratix.emit(...).
No emit site in this bundle bypasses the typed path, so every
emission is tenant-scoped by construction.

Test plan

All 3 adapters: uv run mypy --strict passes
All 3 adapters: every emit_dict_event call replaced
(grep -c 'emit_dict_event' = 0)
All 3 adapters: regression test asserts typed payloads only
All 3 adapters: regression test asserts no DeprecationWarning
All 3 adapters' existing test suites pass (46/46)
Reviewer: confirm canonical model field paths match adapter
emission shape on the wire
Reviewer: confirm ms_agent_framework agent.state.change
collapse is the right call (vs. emitting a synthetic state-
change with hashed metadata)

Bundle progression

Bundle Feat | LAY-874 unit tests #1 (PR feat(instrument): Typed Pydantic event foundation + agno reference (1/17 adapters) #129) — foundation + agno (1 adapter, merged target)
Bundle docs | LAY-881 Initial version of the SDK docs for gitbooks #2 (PR feat(instrument): Typed Pydantic events — autogen + crewai + smolagents (3 adapters / ~23 sites) #138) — autogen + crewai + smolagents (3 adapters / 23 sites)
Bundle docs | LAY-881 Fix wrong yaml structure #3 (this PR) — google_adk + llama_index + ms_agent_framework (3 adapters / 35 sites)
Bundle docs | LAY-881 Added capability to run the gitbook sync manually for … #4 (PR open) — bedrock_agents + openai_agents
Bundle docs | LAY-881 Removed github action. Updated SUMMARY.md to test inte… #5 — TBD
Bundle Improve UX in SDK #6 — TBD

Migrate all 11 emission sites in google_adk lifecycle.py from emit_dict_event(...) to emit_event(TypedModel.create(...)) using the canonical Pydantic models from layerlens.instrument._compat.events (PR #129 foundation). Sites migrated (grep-counted, not estimated): - _before_agent_callback -> AgentInputEvent - _after_agent_callback -> AgentOutputEvent - _after_model_callback -> ModelInvokeEvent + CostRecordEvent - _after_tool_callback -> ToolCallEvent - on_agent_start -> AgentInputEvent - on_agent_end -> AgentOutputEvent - on_handoff -> AgentHandoffEvent (sha256:<hex64> format) - on_tool_use -> ToolCallEvent - on_llm_call -> ModelInvokeEvent - _emit_agent_config -> EnvironmentConfigEvent Total: 11 emission sites (matches estimate). ADK-specific provenance (framework, agent_name, timestamp_ns, session_id, description, instruction) is carried on canonical metadata / attributes / parameters slots. Helpers added: - _stringify: coerce ADK Content/dict/None to canonical message string - _sha256_of: produce sha256:<hex64> for handoff context hash - _coerce_to_dict: wrap scalar tool input/output in {value: ...} - _detect_provider: derive provider from model identifier (Gemini default) Regression tests added: - test_google_adk_lifecycle_emits_typed_payloads_only — every emission is a Pydantic model instance, not a dict - test_google_adk_emit_does_not_warn_after_migration — no DeprecationWarning fires from any lifecycle path mypy --strict: clean (2 source files) pytest: 15/15 passing

Migrate all 12 emission sites in llama_index lifecycle.py from emit_dict_event(...) to emit_event(TypedModel.create(...)) using the canonical Pydantic models from layerlens.instrument._compat.events (PR #129 foundation). Sites migrated (grep-counted, not estimated): - _on_llm_end -> ModelInvokeEvent + CostRecordEvent (when usage) - _on_tool_call -> ToolCallEvent - _on_retrieval_end -> ToolCallEvent (name=retrieval, library) - _on_agent_step_start -> AgentInputEvent - _on_agent_step_end -> AgentOutputEvent - on_agent_start -> AgentInputEvent - on_agent_end -> AgentOutputEvent - on_tool_use -> ToolCallEvent - on_llm_call -> ModelInvokeEvent - on_handoff -> AgentHandoffEvent (sha256:<hex64> format) - _emit_agent_config -> EnvironmentConfigEvent Total: 12 emission sites (matches estimate). LlamaIndex-specific provenance (framework, agent_name, step, timestamp_ns, tool_type, result_count) is carried on canonical metadata / attributes / parameters / input slots. Retrieval events map onto ToolCallEvent with name='retrieval' and integration=IntegrationType.LIBRARY — the canonical schema has no dedicated retrieval shape, mirroring agno reference adapter. Helpers added: _stringify, _coerce_to_dict, _sha256_of, _detect_provider. Regression tests added: - test_llama_index_lifecycle_emits_typed_payloads_only - test_llama_index_emit_does_not_warn_after_migration mypy --strict: clean (2 source files) pytest: 15/15 passing

Migrate all 12 emission sites in ms_agent_framework lifecycle.py from emit_dict_event(...) to emit_event(TypedModel.create(...)) using the canonical Pydantic models from layerlens.instrument._compat.events (PR #129 foundation). Sites migrated (grep-counted, not estimated): - _process_message handoff -> AgentHandoffEvent - _process_message FunctionCall -> ToolCallEvent (input) - _process_message FunctionResult -> ToolCallEvent (output) - _process_message model metadata -> ModelInvokeEvent - _process_message usage metadata -> CostRecordEvent - on_run_start -> AgentInputEvent - on_run_end (output) -> AgentOutputEvent - on_run_end (state.change) -> COLLAPSED into AgentOutputEvent metadata - on_tool_use -> ToolCallEvent - on_llm_call -> ModelInvokeEvent - on_handoff -> AgentHandoffEvent (sha256:<hex64> format) - _emit_chat_config -> EnvironmentConfigEvent Total: 12 emission sites (matches estimate). Behavioural change: the previous adapter emitted an ad-hoc agent.state.change payload alongside agent.output to carry a run_complete / run_failed marker. That payload did not satisfy the canonical AgentStateChangeEvent before_hash / after_hash contract (the run boundary has no real state mutation to hash). The post- migration mapping carries the same signal as run_status on the AgentOutputEvent metadata, preserving the cross-cutting completion marker without violating the canonical schema. Test coverage updated to assert agent.state.change is no longer emitted and run_status is on AgentOutputEvent.metadata. MS Agent Framework-specific provenance (framework, agent_name, chat_name, chat_type, timestamp_ns, selection_strategy, termination_strategy, plugins) is carried on canonical metadata / attributes / parameters / input slots. Helpers added: _stringify, _coerce_to_dict, _sha256_of. Regression tests added: - test_ms_agent_framework_lifecycle_emits_typed_payloads_only - test_ms_agent_framework_emit_does_not_warn_after_migration mypy --strict: clean (2 source files) pytest: 16/16 passing

mmercuri added 3 commits May 10, 2026 10:39

mmercuri requested a review from m-peko May 10, 2026 18:01

This was referenced May 10, 2026

feat(instrument): typed events Bundle #5 — pydantic_ai + semantic_kernel + strands #153

Closed

feat(instrument): typed events Bundle #6 (final) — _base + langchain + langgraph + agentforce + protocols #154

Closed

m-peko closed this May 21, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(instrument): typed events Bundle #3 — google_adk + llama_index + ms_agent_framework#151

feat(instrument): typed events Bundle #3 — google_adk + llama_index + ms_agent_framework#151
mmercuri wants to merge 3 commits into
feat/instrument-typed-events-foundationfrom
feat/instrument-typed-events-bundle-3-google-adk-llama-index-ms-agent-framework

mmercuri commented May 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

mmercuri commented May 10, 2026

Summary

Honest site counts (grep, not estimated)

Per-adapter status

google_adk (commit ffa1aca)

llama_index (commit 4fc3a29)

ms_agent_framework (commit 0917481)

Combined verification

Multi-tenancy compliance

Test plan

Bundle progression

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

google_adk (commit `ffa1aca`)

llama_index (commit `4fc3a29`)

ms_agent_framework (commit `0917481`)