# Changelog

> **This is the PUBLIC form of this artifact. Findings are listed in full, but the details of SECURITY findings — which rule fired, in which file, on which line, and how to fix it — are deliberately withheld, and any secret-scanner results are excluded entirely. Where detail is absent here it was REMOVED FOR PUBLICATION; it is not missing from the analysis. The complete artifact is available from the repository owner.**

## Score

- CAI 52.3 → 57.0 (+4.7)
- Rubric changed (rubric-2026.11.2 → rubric-2026.11.4) — scores are not directly comparable.

## Lenses

- Code Health 63.2 → 61.4 (-1.8)
- Architecture 65.5 → 70.5 (+5.0)
- Maturity 82.4 → 82.0 (-0.4)
- Readiness 56.7 → 73.5 (+16.8)
- Security 76.3 → 82.3 (+6.0)
- Accessibility 38.8 → 43.1 (+4.3)
- Performance 94.3 → 94.3 (0.0)

## Resolved (1183)

- AGUIAdapter._dump_request_parts (cognitive 59) (pydantic_ai_slim/pydantic_ai/ui/ag_ui/_adapter.py)
- AGUIAdapter._dump_request_parts (cyclomatic 29) (pydantic_ai_slim/pydantic_ai/ui/ag_ui/_adapter.py)
- AGUIAdapter._dump_response_parts (cognitive 52) (pydantic_ai_slim/pydantic_ai/ui/ag_ui/_adapter.py)
- AGUIAdapter._dump_response_parts (cyclomatic 26) (pydantic_ai_slim/pydantic_ai/ui/ag_ui/_adapter.py)
- AGUIAdapter.load_messages (cognitive 148) (pydantic_ai_slim/pydantic_ai/ui/ag_ui/_adapter.py)
- AGUIAdapter.load_messages (cyclomatic 66) (pydantic_ai_slim/pydantic_ai/ui/ag_ui/_adapter.py)
- AbstractAgent._infer_name (cognitive 22) (pydantic_ai_slim/pydantic_ai/agent/abstract.py)
- AbstractAgent.run_stream (cognitive 32) (pydantic_ai_slim/pydantic_ai/agent/abstract.py)
- AbstractAgent.run_stream (cyclomatic 20) (pydantic_ai_slim/pydantic_ai/agent/abstract.py)
- Agent._build_toolset_list (cognitive 16) (pydantic_ai_slim/pydantic_ai/agent/__init__.py)
- Agent._open_realtime_session (cognitive 19) (pydantic_ai_slim/pydantic_ai/agent/__init__.py)
- Agent._open_realtime_session (cyclomatic 20) (pydantic_ai_slim/pydantic_ai/agent/__init__.py)
- Agent._prepare_run (cognitive 139) (pydantic_ai_slim/pydantic_ai/agent/__init__.py)
- Agent._prepare_run (cyclomatic 100) (pydantic_ai_slim/pydantic_ai/agent/__init__.py)
- Agent._resolve_realtime_session (cognitive 45) (pydantic_ai_slim/pydantic_ai/agent/__init__.py)
- Agent._resolve_realtime_session (cyclomatic 38) (pydantic_ai_slim/pydantic_ai/agent/__init__.py)
- Agent.override (cognitive 69) (pydantic_ai_slim/pydantic_ai/agent/__init__.py)
- Agent.override (cyclomatic 48) (pydantic_ai_slim/pydantic_ai/agent/__init__.py)
- AgentStream.validate_response_output (cognitive 26) (pydantic_ai_slim/pydantic_ai/result.py)
- AgentStream.validate_response_output (cyclomatic 22) (pydantic_ai_slim/pydantic_ai/result.py)
- …and 1163 more

## New (40)

- Change-coupling hub: app.py → api_keys.py, sessions.py, model_menu.py, status.py (src/pydantic_clai2/pydantic_clai2/app.py)
- Documentation: no build, run or usage example (pydantic_ai_slim/README.md)
- High: security finding (details withheld)
- High: security finding (details withheld)
- High: security finding (details withheld)
- High: security finding (details withheld)
- No ADRs found (pyproject.toml)
- Off-boarding risk: anonymized user #2
- Off-boarding risk: anonymized user #1
- Orphaned file (pydantic_ai_slim/pydantic_ai/models/_reasoning_details.py)
- Orphaned file (pydantic_ai_slim/pydantic_ai/models/crusoe.py)
- Orphaned file (pydantic_ai_slim/pydantic_ai/ui/vercel_ai/response_types.py)
- Orphaned file (src/pydantic_ai_harness/pydantic_ai_harness/browser_use/_settings.py)
- Orphaned file (src/pydantic_ai_harness/pydantic_ai_harness/planning/_types.py)
- Test declared unreliable: test_code_mode_runs_remotely (tests/harness/code_mode/test_remote.py)
- Test declared unreliable: test_code_mode_runs_remotely_in_temporal_workflow (tests/harness/code_mode/test_temporal.py)
- Test declared unreliable: test_disconnect_mid_snippet_reports_started_calls (tests/harness/code_mode/test_remote.py)
- TodoComment (pydantic_ai_slim/pydantic_ai/capabilities/__init__.py)
- TodoComment (pydantic_ai_slim/pydantic_ai/capabilities/image_generation.py)
- TodoComment (pydantic_ai_slim/pydantic_ai/capabilities/native_or_local.py)
- …and 20 more

## Changes since last survey

- 300 commits — 288 feature/other, 12 fixes

## By area

- pydantic_ai_slim/pydantic_ai — 87 commits
- src/pydantic_ai_harness — 59 commits
- src/pydantic_clai2 — 27 commits
- .github/scripts — 17 commits
- tests/harness — 17 commits
- .github/workflows — 14 commits
- tests/models — 12 commits
- (root) — 6 commits
- tests/clai2 — 5 commits
- .agents/skills — 4 commits
- tests/durable_exec — 4 commits
- .github/api-compatibility-waivers — 3 commits
- docs/evals — 3 commits
- docs/harness — 3 commits
- docs/models — 3 commits
- pydantic_evals/pydantic_evals — 3 commits
- tests/realtime — 3 commits
- docs/capabilities — 2 commits
- docs/comparisons — 2 commits
- docs/graph.md — 2 commits

## Notable commits

- fix: Fix 16 CI flakes in the tests, so a runner stall no longer changes their outcome (#12933)
- fix: Fix `HasMatchingSpan` docs recipes to check every span in the trace (#13011)
- fix: Fix `MCPSamplingModel` system prompt separator and conditional `<system>` announcement (#12488)
- fix: Fix `OpenAIChatModel` sending Responses reasoning item ids as assistant-message fields in `auto` thinking mode (#12136)
- fix: Fix `TestModel` tool-args generation for schemas with list-form `type` (#13097)
- fix: Fix broken Render workspace and Keenable Search links in the docs (#13046)
- fix: Fix flaky `RenderSandbox` supervisor output-limit test (#13454)
- fix: Fix flaky `test_remembered_approval_holds_on_replay_and_in_the_next_run` by closing the event loop `run_sync` leaves on the DBOS executor thread (#13319)
- fix: Fix flaky `test_run_sync_token_cancel_from_another_thread` by closing the event loop `run_sync` leaves on its worker thread (#13566)
- fix: Fix flaky `test_stopping_a_command_kills_its_descendants` by treating a descendant reaped mid-read as gone (#13385)
- fix: Fix flaky `test_timeout_interrupts_code_that_keeps_printing` by starting its timeout once the cell runs (#13353)
- fix: Fix the `test_each_browser_gets_only_the_sessions_it_watches` flake by waiting for the queued input it asserts on (#13609)
- change: Accept a UTF-8 byte order mark in `Dataset.from_file` and `AgentSpec.from_file` (#8317)
- change: Accept an async `api_key` callable in `OpenAIProvider` (#7886)
- change: Accept prose before a bare JSON object in douwebot's OpenCode answer, and keep a rejected answer out of the job log (#13286)
- change: Accept prose before the JSON object in every CI OpenCode answer, including prose that holds a `{` (#13305)
- change: Accept prose before the closing JSON fence in `PR Review Checks` answers (#13220)
- change: Add Claude Fable 5.2 (`claude-fable-5-2`) and Claude Mythos 5.2 (`claude-mythos-5-2`) support (#12970)
- change: Add GPT-6.1 Terra to the Bedrock Converse profile gate (#12731)
- change: Add `BoxdSandbox(secrets=)`, naming the boxd secrets a new machine gets (#13589)
- …and 280 more

## Architecture

- Containers 0 added · 0 removed · contexts 0 added · 1 removed · edges 0 added · 0 removed

## Removed bounded contexts (1)

- pydantic-ai
