Every commit scored for change-risk against this repo's own history, so 'elevated' means elevated here rather than on some global curve.
Needs review
167 commits sit in this repo's top risk tercile, which is 33% of the 500scored. The cut is drawn against this codebase's own history rather than a global curve, so a quiet repo still fills its top band, and here it starts at 8.3 out of 10. What pushes a commit up is size and spread together: a large change confined to one area scores below a smaller one scattered across a dozen files.
Commit categories over time, read off the subject line. Fixes carry the accent because that is the series this chart exists to show.
Consistently fix-driven across its history.
Ranked by change-risk, highest first. Priority is a tercile of this repo's own distribution, so a quiet repo still fills its top band.
| # | Commit | Author | When | Lines | Risk | Top driver |
|---|---|---|---|---|---|---|
| 1 | 07aa1e09fix(live): keep streaming tool yields from completing turns | Yufeng He | 1mo ago | +478.0K -0 | 100%Elevated | more lines added than baseline |
| 2 | 456524d7test: add unit tests for public symbols that had no coverage | George Weale | 5d ago | +13.6K -1 | 99%Elevated | more lines added than baseline |
| 3 | 42f220a6fix(models): report a missing Anthropic credential when the client is built | George Weale | 5d ago | +10.1K -3 | 99%Elevated | more lines added than baseline |
| 4 | 467d972ctest(telemetry): make the functional tests record/replay | Max Ind | 1w ago | +9.3K -6.0K | 99%Elevated | more lines added than baseline |
| 5 | 9adf0113fix(plugins): complete BigQuery Agent Analytics privacy and shutdown hardening | Haiyuan Cao | 2w ago | +5.7K -591 | 99%Elevated | more lines added than baseline |
| 6 | 930e4725refactor: add a2a-sdk 0.3.x/1.x compatibility layer for ADK A2A support | Google Team Member | 4w ago | +4.4K -2.6K | 99%Elevated | more lines added than baseline |
| 7 | 44d747edchore: update compiled adk web assets | adk-bot | 1mo ago | +6.6K -8.7K | 99%Elevated | more lines added than baseline |
| 8 | 04b8b727feat(plugins): add BigQuery Agent Analytics delivery and termination observability | Haiyuan Cao | 1d ago | +2.0K -569 | 98%Elevated | more lines added than baseline |
| 9 | 625ef1aafeat(integrations): add OCI Generative AI provider | Fede Kamelhar | 2w ago | +2.8K -0 | 98%Elevated | more lines added than baseline |
| 10 | d4f157d2feat(eventarc): add Eventarc Advanced toolset for ADK | Google Team Member | 2w ago | +2.9K -0 | 98%Elevated | more lines added than baseline |
| 11 | 2919bf5bfix(plugins): harden BigQuery agent analytics against fail-open privacy, GCS concurrency, and startup-loss gaps | Haiyuan Cao | 3w ago | +2.0K -97 | 98%Elevated | more lines added than baseline |
| 12 | d496a3b3test(telemetry): report token usage in the functional telemetry scenario | Google Team Member | 6d ago | +1.2K -20 | 97%Elevated | more lines added than baseline |
| 13 | 8930d9b1feat(telemetry): add request-driven metric export for Agent Engine | Max Ind | 1w ago | +1.6K -74 | 97%Elevated | more lines added than baseline |
| 14 | f1a0a148feat: Add bigquery-graph skill to ADK BigQuery tools | Google Team Member | 3w ago | +1.1K -0 | 97%Elevated | more lines added than baseline |
| 15 | 327afeb4feat: support audio generation user simulator - Add new LlmAudioUserSimulator which synthesizes text and generates audio along with the text - On top of existing BaseUserSimulatorConfig, accepts: - audio_model: TTS/audio model name - audio_model_configuration: includes a speech_config for defining voice persona and language - include_text_with_audio: whether to final user message should contain both text with the generated audio, or just the audio tokens - LlmAudioUserSimulator supports au | Google Team Member | 3w ago | +1.6K -24 | 97%Elevated | more lines added than baseline |
| 16 | 9d306f5dfeat(workflow): Support state-based resumption for task-mode agent workflow nodes | Shangjie Chen | 4w ago | +1.9K -177 | 97%Elevated | more lines added than baseline |
| 17 | 7d0ae63afeat(plugins): add on_agent_error_callback and on_run_error_callback | Haiyuan Cao | 4w ago | +1.6K -128 | 97%Elevated | more lines added than baseline |
| 18 | 2280f1ccfeat: add telemetry metrics collection for ADK CLI execution | Lucas Kang | 2w ago | +830 -0 | 96%Elevated | more lines added than baseline |
| 19 | 322f4559feat: Add ReflectAndRetryModelPlugin for self-healing model errors | Jason Zhang | 2w ago | +1.2K -40 | 96%Elevated | more lines added than baseline |
| 20 | c0fa51b3feat: add skill to verify markdown code snippets | Shahin Saadati | 4w ago | +1.1K -5 | 96%Elevated | more lines added than baseline |
| 21 | 1263ed64feat: Implement Workflow as Tool core feature | Shangjie Chen | 1mo ago | +1.5K -4 | 96%Elevated | more lines added than baseline |
| 22 | e0b6c9a9chore: check the release wheel imports no worse than the last release | George Weale | 5d ago | +1.1K -0 | 95%Elevated | more lines added than baseline |
| 23 | 4cab3ac1fix: keep regular-tool FRs on mixed task turns | Aarav Mittal | 5d ago | +730 -126 | 95%Elevated | more lines added than baseline |
| 24 | 7f82142arefactor(types): make google.adk.models pass strict mypy | George Weale | 1w ago | +843 -357 | 95%Elevated | more lines added than baseline |
| 25 | 07add3b8refactor(types): make the agents, sessions, flows and workflow packages pass strict mypy | George Weale | 1w ago | +1.2K -635 | 95%Elevated | more lines added than baseline |
| 26 | 73ecb5b5fix: wire App plugins through eval paths | doug | 1w ago | +648 -23 | 95%Elevated | more lines added than baseline |
| 27 | 5091f0a6feat(eval): Make live and audio evals reachable via public entrypoints | Stephen Allen | 2w ago | +951 -101 | 95%Elevated | more lines added than baseline |
| 28 | fa31b6cafeat(environment): Add support for executing skill scripts within an Environment | Liang Wu | 1w ago | +904 -45 | 94%Elevated | more lines added than baseline |
| 29 | 550189cefix: wrap input_schema payload in ReAct prompt and propagate tool_choice to LiteLLM | Eva | 2w ago | +572 -29 | 94%Elevated | more lines added than baseline |
| 30 | 8d2e9d1cchore: Refactored user simulator config to allow for different types of configs | Ankur Sharma | 4w ago | +496 -13 | 94%Elevated | more lines added than baseline |
| 31 | c5672030fix: Secure and harden FileArtifactService against tampered metadata and partial writes | Xuan Yang | 5d ago | +492 -78 | 94%Elevated | more lines added than baseline |
| 32 | 6396ce60fix(telemetry): mark failed tool spans as errors and fix provider naming | George Weale | 1w ago | +589 -22 | 94%Elevated | more lines added than baseline |
| 33 | 27548e39fix: kill runaway code on timeout in container and local executors | George Weale | 1w ago | +523 -3 | 94%Elevated | more lines added than baseline |
| 34 | 217a90a2feat(eventarc): add Eventarc Advanced toolset for ADK | Google Team Member | 2w ago | +682 -300 | 94%Elevated | more lines added than baseline |
| 35 | fa884e3dfix: Update GCP Skill Registry to use agentregistry/skill endpoint | Kathy Wu | 4w ago | +429 -114 | 94%Elevated | more lines added than baseline |
| 36 | 61ddc5fafeat(skills): support non-blocking skill loading in async runtimes | Godwin Paul Vincent | 5d ago | +412 -0 | 92%Elevated | more lines added than baseline |
| 37 | b34c636frefactor(types): make google.adk.a2a pass strict mypy | George Weale | 6d ago | +466 -206 | 92%Elevated | more lines added than baseline |
| 38 | 25101412refactor(types): make google.adk.evaluation pass strict mypy | George Weale | 1w ago | +379 -199 | 92%Elevated | more lines added than baseline |
| 39 | 2aff82c3feat: Introduce a capability reporting system for LLM models | Xuan Yang | 1w ago | +469 -7 | 92%Elevated | more lines added than baseline |
| 40 | 8806dc2bperf: improve adk import loading | George Weale | 1w ago | +515 -118 | 92%Elevated | more lines added than baseline |
| 41 | 6bab08fcfeat: add telemetry consent check, status commands, and interrupt safety to CLI | Lucas Kang | 2w ago | +417 -1 | 92%Elevated | more lines added than baseline |
| 42 | 5a12ee09fix(a2a): Promote RemoteA2aAgent response to workflow node output | Harineko0 | 2w ago | +417 -0 | 92%Elevated | more lines added than baseline |
| 43 | c03f3337fix: Prevent continuation forgery in tool confirmation | Kathy Wu | 3w ago | +431 -196 | 92%Elevated | more lines added than baseline |
| 44 | 45a77dc5fix: Validate path segments in GcsArtifactService and InMemoryArtifactService to prevent cross-user artifact access | Ashutosh0x | 4w ago | +406 -67 | 92%Elevated | more lines added than baseline |
| 45 | 9cd59753feat: let McpToolset reuse the MCP server's tool list | Kathy Wu | 1d ago | +379 -24 | 90%Elevated | more lines added than baseline |
| 46 | cd36dbc3refactor(types): type the integrations, skills and MCP tool packages for strict mypy | George Weale | 4d ago | +487 -227 | 90%Elevated | more lines added than baseline |
| 47 | 0fcb7f15refactor(types): make the code executors, planners and runner pass strict mypy | George Weale | 1w ago | +426 -199 | 90%Elevated | more lines added than baseline |
| 48 | 76d5723ffix: Stop re-validating already-consumed tool confirmations | Xuan Yang | 1w ago | +365 -42 | 90%Elevated | more lines added than baseline |
| 49 | 93db97dbfeat: add --extra_packages option to `adk deploy agent_engine` | George Weale | 2w ago | +465 -2 | 90%Elevated | more lines added than baseline |
| 50 | 3dd1156cfeat(a2a): support per-invocation auth headers when fetching agent cards | Google Team Member | 2w ago | +463 -21 | 90%Elevated | more lines added than baseline |
Two views of the same model: where the cuts fall, and what commit shape lands you above them.
Every scored commit, binned on the raw 0 to 10 score rather than the percentile. Percentile ranks are uniform by construction, so that axis has no shape to draw. The dashed lines are the tercile cuts behind each row's priority pill.
The 200 most recent commits, on their own recency sample rather than the feed above: that defaults to risk-sorted, so reusing it would plot only the top tercile and call it the spread. Big and scattered is what the model penalises. Click a dot to open it.
google/adk-python has 500 commits in its history from 100 contributors, the first of them Jul 6, 2026. In the last 90 days 1,871 files were touched, 3,603 times in total, most often cli_tools_click.py. Every commit is scored for change risk from its size, spread and the history of the files it touches.
