pydantic/pydantic-ai
52.3
Adequate · 5 November 2026
313k
lines of production code
Python
primary language
4
measurements over time
What this system is
Features
Add DBOS integration for durable agent execution
Users can now run Pydantic AI agents as durable workflows on DBOS. This release introduces the \DBOSAgent\ wrapper (now deprecated in favor of the \DBOSDurability\ capability) and the \DBOSDurability\ capability itself, which offload model requests, tool calls, and MCP server communication to checkpointed DBOS steps. The integration supports configurable step settings for model and tool operations, enforces unique IDs for MCP toolsets to prevent caching collisions, and provides a \DBOSOperationNamer\ to ensure stable step names for workflow recovery. It also includes utilities like \StepConfig\ for configuring retries and backoff, and guards against unsafe operations like \ctx.enqueue()\ inside durable steps.
_pydantic\_ai\_slim/pydantic\_ai/durable\exec/dbos · high confidence
Add LangChain community tools integration
Users can now easily use LangChain community tools in PydanticAI. The new \pydantic\_ai.ext.langchain\ module provides \LangChainToolset\ and \tool\_from\_langchain\ to wrap LangChain tools as PydanticAI tools, allowing seamless integration of existing LangChain tooling into PydanticAI agents.
_pydantic\_ai\_slim/pydantic\ai/ext · high confidence
Add PromptInjectionDefender capability for indirect prompt injection protection
Users can now protect their agents from indirect prompt injection attacks by adding the \PromptInjectionDefender\ capability to their Pydantic AI agents. This new feature inspects untrusted text returned by tools (such as emails, documents, or web content) and can block high-risk results or trigger custom callbacks. It supports pattern-based detection by default and optional ML-based semantic detection via an extra package, allowing developers to configure blocking thresholds, custom defense providers, and detection logging.
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/prompt\_injection\defender · high confidence
Add Slack Lead Qualifier example
Introduces a new example application that demonstrates an AI-driven workflow for qualifying new Slack members. The system uses a Pydantic AI agent (powered by OpenAI) to research new joiners via DuckDuckGo, scoring their relevance to Pydantic Logfire. It includes a FastAPI webhook to receive Slack team-join events, Modal serverless functions to handle asynchronous processing and daily summaries, and a Modal Dict store to persist analysis results.
_examples/pydantic\_ai\_examples/slack\_lead\qualifier · high confidence
Add You.com web search and research capabilities to the harness
The harness now includes \YouSearch\ and \YouResearch\ capabilities, enabling agents to perform web research using the You.com API. \YouSearch\ provides \web\_search\ (returning excerpts or full pages) and \get\_page\ tools for surveying and reading sources, while \YouResearch\ adds \answer\, \research\, and \finance\_research\ tools for multi-step, cited answers. These capabilities support configuration for domain filtering, result freshness, and native search fallback, and integrate with durable execution to record API requests.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/youdotcom · high confidence
Add testing skill and cassette parsing utility
Introduces a new testing skill for the AI agent that guides users through recording, verifying, and debugging VCR cassettes. This includes a new \parse\_cassette.py\ script that reads cassettes (supporting both the \cassetter\ format and legacy VCR.py formats) and pretty-prints request/response bodies with truncated base64 strings for readability, alongside updated documentation in \SKILL.md\ detailing the pytest workflow and required flags.
.claude/skills/testing-skill · high confidence
Experimental ACP adapter for editor integration
Adds an experimental \pydantic\_ai\_harness.experimental.acp\ module that exposes a Pydantic AI agent to editors and terminal UIs over the Agent Client Protocol (ACP) via stdio. The new \run\_acp\_stdio\ and \run\_acp\_stdio\_sync\ entry points serve the agent as an ACP subprocess, translating ACP JSON-RPC requests into agent runs. The adapter supports streaming text, human-in-the-loop tool approval (mapping ACP permission requests to deferred-approval tools), and rich tool rendering for \FileSystem\ and \Shell\ calls (annotating reads, edits, and executions with file locations and diffs). It enables per-workspace sessions by rooting tools at the client's working directory and offers optional editor-native filesystem and shell toolsets (\acp\_filesystem\, \acp\_terminal\) that route I/O through the client when advertised, falling back to local capabilities otherwise. The module also handles ACP content block translation (text, images, audio, embedded resources), bounded text chunking to prevent buffer overruns, and MCP server wiring via \session\_config\.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/experimental/acp · high confidence
Expose shared agent skills to Claude Code
Claude Code now automatically discovers and exposes shared agent skills and their associated AGENTS.md guides. This change introduces symbolic links in the .claude/skills directory that point to the central skill definitions located in .agents/skills, making capabilities such as completing partial PRs, auditing features, and managing model additions directly available within the Claude Code environment.
.claude/skills · high confidence
Introduce AbstractAgent and WrapperAgent for extensible agent architectures
The agent module now exposes \AbstractAgent\ as a base class for building custom agent implementations and \WrapperAgent\ as a decorator-style wrapper that delegates to another agent. This allows users to intercept and modify agent behavior, inject custom capabilities, or adapt agents for specific execution environments (such as durable execution engines) without modifying the core \Agent\ class. The \WrapperAgent\ transparently forwards properties like \model\, \name\, \description\, and \toolsets\, while allowing overrides for validation context and event handling, enabling clean composition and extension of agent logic.
_pydantic\_ai\_slim/pydantic\ai/agent · high confidence
Introduce Shell capability for agent command execution
The \pydantic\_ai\_harness\ now includes a \Shell\ capability that allows agents to run shell commands within the run's workspace. This feature provides tools for synchronous execution (\run\_command\) and background process management (\start\_command\, \check\_command\, \stop\_command\), along with an opt-in persistent \shell\ tool. It includes configurable security controls such as allow/deny lists for commands and operators, environment variable filtering, and per-file size limits to prevent resource exhaustion. The capability is designed to handle output truncation, timeouts, and recoverable errors gracefully, enabling agents to perform tasks like building, testing, or linting code.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/shell · high confidence
Introduce StackOne capability for linked SaaS account actions
Added a new StackOne capability that allows agents to interact with linked business applications (such as BambooHR, Salesforce, or Zendesk) via the StackOne MCP endpoint. This includes the \StackOne\ capability class for agent integration, the \StackOneToolset\ implementation for tool execution, and support for two tool modes: \search\_execute\ (default, using search and execute meta-tools) and \individual\ (one tool per action). The feature supports action filtering via glob patterns, requires HTTPS for API connections, and reads credentials from the \STACKONE\_API\_KEY\ environment variable or explicit arguments.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/stackone · high confidence
Introduce Vercel AI protocol adapter for streaming and history round-trips
Adds the \VercelAIAdapter\ and \VercelAIEventStream\ to enable Pydantic AI agents to integrate with the Vercel AI protocol. This adapter handles bidirectional message conversion, supporting streaming event-based communication and history persistence via \dump\_messages\ and \load\_messages\. It targets Vercel AI SDK versions 5, 6, and 7, with SDK 6+ enabling human-in-the-loop tool approval workflows. The implementation preserves provider metadata, file data, and message IDs across round-trips, and includes security measures such as strict boolean validation for tool approvals and sanitization of file URL schemes.
_pydantic\_ai\_slim/pydantic\_ai/ui/vercel\ai · high confidence
Introduce direct image generation API with \`ImageGenerator\`
This change adds a new direct image generation capability to the library, centered on the \ImageGenerator\ interface and the \ImageGenerationModel\ abstract base class. It introduces a portable settings system (\ImageGenerationSettings\) allowing users to specify output dimensions or aspect ratios, along with validation logic to enforce mutual exclusivity between these options and to handle provider-specific geometry preferences. The implementation includes utilities for media type sniffing from binary data, a wrapper model class for extending image generation behavior (such as caching or logging), and input validation for image sources (URLs, binary data, or uploaded files).
_pydantic\_ai\_slim/pydantic\ai/images · high confidence
Introduce durable execution support for Temporal workflows
This change adds a new \pydantic\_ai.durable\_exec.temporal\ module that enables running Pydantic AI agents as durable workflows on the Temporal platform. It provides a \TemporalAgent\ and \TemporalDurability\ capability to offload model requests, tool calls (including \FunctionToolset\, \DynamicToolset\, and \MCPToolset\), and event streaming into Temporal activities. The implementation includes a custom \PydanticAIPayloadConverter\ for serializing complex tool arguments and run contexts, a \LogfirePlugin\ with replay-safe tracing to prevent duplicate spans during workflow replays, and a cancellation-safe activity executor to prevent livelocks when anyio cancel scopes are triggered. It also passes through necessary SDK modules (e.g., \openai\, \anthropic\, \httpx\) to the Temporal sandbox to allow model construction and HTTP calls within activities.
_pydantic\_ai\_slim/pydantic\_ai/durable\exec/temporal · high confidence
Introduce dynamic and external toolset abstractions
The toolsets module now includes DynamicToolset, which allows tool collections to be built or modified based on the run context (with optional per-step re-evaluation), and ExternalToolset, which defines tools whose execution happens outside the agent run. These additions, along with the new module structure and guidelines, enable more flexible, context-aware tool composition and lifecycle management.
_pydantic\_ai\_slim/pydantic\ai/toolsets · high confidence
Introduce embeddings API with provider-specific model implementations
Users can now generate vector embeddings using a new \EmbeddingModel\ abstraction and dedicated implementations for OpenAI, Cohere, Google, Bedrock (including Titan, Cohere, and Nova models), Sentence Transformers, and VoyageAI. The change adds a \settings\ module for common and provider-specific configuration (such as dimensions, truncation, and Bedrock inference profiles) and a \wrapper\ module to allow wrapping models for behaviors like caching or logging.
_pydantic\_ai\_slim/pydantic\ai/embeddings · high confidence
Introduce pydantic\_evals package for LLM evaluation
The new \pydantic\_evals\ package provides a toolkit for evaluating stochastic functions like LLM calls. It introduces \Dataset\ and \Case\ for managing structured test data, \CaseLifecycle\ hooks for setup, context enrichment, and teardown during evaluation, and \generate\_dataset\ for creating test cases via LLMs. The package also supports online evaluation observability by emitting OpenTelemetry \gen\_ai.evaluation.result\ events and includes a \py.typed\ marker for type-checking support.
_pydantic\_evals/pydantic\evals · high confidence
Introduce pydantic\_graph library for type-hint based agent graph execution
The \pydantic\_graph\ package is introduced, providing a library for constructing and executing typed agent graphs. Users can define graph structures using \GraphBuilder\ and typed step functions or \BaseNode\ subclasses, then execute them via \Graph\ or \GraphRun\. The library exposes primitives for node definitions (\StartNode\, \EndNode\, \Step\, \Join\, \Fork\, \Decision\), error handling (\GraphSetupError\, \GraphRuntimeError\), and utility types (\TypeExpression\, \Maybe\). It also includes robust synchronous execution support via \run\_until\_complete\, which handles event loop management and cleanup on interrupts, raising \UnsupportedEventLoopError\ for incompatible loops (e.g., Temporal workflows).
_pydantic\_graph/pydantic\graph · high confidence
Introduce runtime capability creation for agent-authored extensions
Agents can now author, validate, and persist Pydantic AI capabilities during a run, which are then activated in subsequent runs. This new \CapabilityCreation\ feature allows coding agents to dynamically write Python source code for new capabilities (such as tools or guardrails) and have them available without a process restart. The implementation includes a toolset for writing and managing these capabilities, a validation layer that ensures authored code conforms to the \AbstractCapability\ interface, and support for durable execution to persist authoring actions across workflow recoveries.
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/capability\creation · high confidence
Introduce span-tree recording for agent evaluations
The \pydantic\_evals\ package now includes an OpenTelemetry-based span-tree feature, allowing users to capture and analyze the full trace of agent runs (including tool calls and model requests) via the \context\_subtree()\ context manager. This addition exports span data as structured \SpanTree\ objects, enabling new agentic evaluators like \ToolCorrectness\ and \TrajectoryMatch\ to inspect execution paths. The implementation gracefully handles missing dependencies or unsupported tracer providers by yielding a \SpanTreeRecordingError\ rather than crashing, ensuring evaluations continue even if span recording is unavailable.
_pydantic\_evals/pydantic\evals/otel · high confidence
Introduce structured provider API with model profiles and shared HTTP client lifecycle
The \pydantic\ai/providers\ package now exposes a formal \Provider\ base class that owns API clients, authentication, base URLs, and HTTP lifecycle management, along with a \ModelProfile\ system that lets each provider declare model-specific capabilities (such as thinking, structured output, and document input) independent of the model class. This change centralizes HTTP client creation and disposal (including support for \httpx2\ and legacy \httpx\ clients) and provides a shared \OpenAICompatibleProvider\ base for OpenAI-compatible APIs, while adding new provider implementations for Alibaba, Azure, Bedrock Mantle, Cerebras, Cohere, Crusoe, and DeepSeek, each wired into the provider inference system via \\\init\\_.py\.
_pydantic\_ai\_slim/pydantic\ai/providers · high confidence
Introduce the Capabilities system for composable agent behavior
Pydantic AI now uses a new 'Capabilities' abstraction to manage cross-cutting agent behavior, replacing direct constructor arguments with a composable list of capabilities. This change introduces a registry of built-in capabilities—including NativeTool, WebSearch, ImageGeneration, ProcessHistory, Instrumentation, and Thinking—that can be composed to configure agent behavior. The system supports dynamic capabilities, event handling via @on\_event, and specific ordering constraints to ensure correct composition. Existing patterns like fallback\_model are deprecated in favor of capability-based configuration (e.g., fallback\_subagent\_model).
_pydantic\_ai\_slim/pydantic\ai/capabilities · high confidence
Introduction of experimental capabilities module with warning system
A new \pydantic\_ai\_harness.experimental\ package has been added to host experimental features that may change or be removed without a deprecation period. Importing specific experimental capabilities now emits a \HarnessExperimentalWarning\ to alert users of their unstable nature, while importing the \HarnessExperimentalWarning\ class itself allows users to silence these warnings globally. The module also includes a \warn\_moved\ helper to guide users when capabilities graduate out of the experimental namespace to the top-level package.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/experimental · high confidence
Introduction of pydantic-evals package with agentic and report evaluators
The \pydantic\_evals\ package is now available, providing a comprehensive suite of evaluators for assessing AI agent performance. This release introduces agentic-specific evaluators such as \ToolCorrectness\, \TrajectoryMatch\, and \ArgumentCorrectness\ to analyze tool usage and decision paths, alongside report-level evaluators like \ROCAUCEvaluator\ and \ConfusionMatrixEvaluator\ for statistical analysis. The package also includes standard quality metrics via \GEval\ and \LLMJudge\, while removing the insecure \Python\ evaluator for safety reasons.
_pydantic\_evals/pydantic\evals/evaluators · high confidence
Introduction of the \`clai\` CLI tool
The \clai\ package is now available as a command-line interface for interacting with LLMs. It supports interactive chat sessions, one-shot prompts, and a web-based chat UI (launchable via \clai web\). Users can specify models (defaulting to \openai:gpt-5\), custom agents, code themes, and MCP server configurations. The package includes its own MIT license and documentation.
clai · high confidence
New AG-UI API examples for agent interaction patterns
The \examples/pydantic\_ai\_examples/ag\_ui/api\ directory now provides a set of Starlette-based API examples demonstrating various agent interaction patterns using the \AGUIAdapter\. These include agentic chat, generative UI with plan steps, human-in-the-loop workflows, predictive state updates, shared state management, tool approval interrupts, and tool-based generative UI. Each example exposes a POST endpoint that dispatches requests to a Pydantic AI agent configured with specific tools, instructions, and state dependencies to showcase how the AG-UI protocol handles different agent behaviors.
_examples/pydantic\_ai\_examples/ag\ui/api · high confidence
New AG-UI example application with multiple demo endpoints
A new example application has been added to demonstrate the Pydantic AI agent integrated with the AG-UI protocol. This FastAPI-based server exposes seven distinct endpoints for users to explore various agentic capabilities, including Agentic Chat, Human in the Loop, Agentic Generative UI, Tool Based Generative UI, Shared State, Predictive State Updates, and Tool Approval (interrupts). The application can be run locally via a CLI command using Uvicorn on port 9000.
_examples/pydantic\_ai\_examples/ag\ui · high confidence
New Researcher capability for web research agents
The \Researcher\ capability is now available in the Pydantic AI Harness, providing a pre-configured stack for broad web research with source-backed answers. It combines \WebSearch\ (with DuckDuckGo fallback), \WebFetch\, \SubAgents\ for delegation, and \ToolOutputLimits\ to spill oversized results to the workspace. Users can instantiate it via \Researcher()\ alongside other capabilities or run the exported \researcher\_agent\ directly via the CLI.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/researcher · high confidence
New UI adapter module with strict content-type enforcement and backward-compatible event streaming
The \pydantic\_ai.ui\ package introduces a new UI adapter layer that enforces \application/json\ as the default \Content-Type\ for inbound requests to prevent cross-origin CSRF attacks, while maintaining backward compatibility with older AG-UI versions through version-gated functionality and graceful skipping of unknown message types. The module provides \UIAdapter\ and \UIEventStream\ base classes that manage message building, event encoding, and state handling, ensuring that AG-UI response identities are correctly tracked across tool calls and that streaming errors properly close open message parts.
_pydantic\_ai\_slim/pydantic\ai/ui · high confidence
New and updated example applications for Pydantic AI
The examples directory now includes several new demonstration applications: a CLI tool to copy examples to a new directory, an interactive chat app demonstrating how to cancel and resume streaming agent runs, a browser-based chat interface (with TypeScript transpiled in-browser), a realtime camera and voice assistant example, a simple realtime text-to-audio example, a realtime voice assistant with barge-in support, a markdown streaming example, a structured response streaming example, and a Gradio-based weather agent. Additionally, the py.typed marker file has been moved from the pydantic\_ai package root to the examples directory.
_examples/pydantic\_ai\examples · high confidence
New evals examples for time-range inference agents
The \examples/pydantic\_ai\_examples/evals\ directory now provides a complete workflow for evaluating a time-range inference agent. It includes a generated dataset (\time\_range\_v1.yaml\) and a schema (\time\_range\_v1\_schema.json\) for test cases, along with scripts to generate datasets, add custom evaluators (like \ValidateTimeRange\ and \AgentCalledTool\), and run evaluations against different models (e.g., \openai:gpt-5.1\ vs \openai:gpt-5.2\).
_examples/pydantic\_ai\examples/evals · high confidence
New media externalization subsystem for step persistence
The \pydantic\_ai\_harness.media\ package introduces content-addressed storage plumbing to keep step-persistence snapshots small by offloading large binary and text payloads. It provides \DiskMediaStore\, \SqliteMediaStore\, \S3MediaStore\, and a new \MongoMediaStore\ (available via the \mongodb\ extra) that store payloads keyed by SHA-256 hash, enabling deduplication across messages and snapshots. The \externalize\_media\ and \restore\_media\ helpers walk message trees to replace large \BinaryContent\ and text parts (exceeding a configurable byte threshold) with \media+sha256://\ URIs, and rehydrate them on demand. This change is a behavioral improvement to the persistence layer, reducing snapshot size without altering the agent's functional capabilities. Also in this area: PostgresStepStore and PostgresMediaStore for hosted PostgreSQL persistence.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/media · high confidence
New realtime WebRTC voice agent example
Added a new example demonstrating a browser-based voice agent that exchanges audio directly with OpenAI over WebRTC for lowest latency, while a Pydantic AI server-side sideband handles tools and history. The example includes a FastAPI backend, an HTML/JS frontend for microphone access and call control, and documentation explaining the topology and setup.
_examples/pydantic\_ai\_examples/realtime\webrtc · high confidence
Prefect durable execution integration
Pydantic AI now supports durable execution with Prefect, allowing agent runs to survive infrastructure failures and resume automatically. This location provides the Prefect-specific integration layer, including the \PrefectAgent\ wrapper (now deprecated in favor of the \PrefectDurability\ capability) and dedicated toolset wrappers (\PrefectFunctionToolset\, \PrefectMCPToolset\, and dynamic toolset support) that offload model requests, tool calls, and MCP server communication to Prefect tasks. The integration includes a custom cache policy (\DEFAULT\_PYDANTIC\_AI\_CACHE\_POLICY\) and key projection logic to ensure that tool results are correctly replayed on flow retries without colliding on non-deterministic \RunContext\ fields, while also enforcing non-retryable conditions for specific framework and workspace errors.
_pydantic\_ai\_slim/pydantic\_ai/durable\exec/prefect · high confidence
WarnOnCacheBusts capability detects prompt cache collapses across conversation turns
The \WarnOnCacheBusts\ capability now monitors prompt-cache health by tracking cache hit ratios across consecutive requests within a run and across runs that continue the same conversation. It emits \CacheBustWarning\ when the cacheable prefix collapses unexpectedly (e.g., due to reordered tools or injected timestamps) and \CacheNotEnabledWarning\ when caching is not configured for models that require it. The monitor classifies collapses against provider retention windows to distinguish between moved prefixes and expired caches, and it handles native tool calls and model switches to avoid false positives.
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/warn\_on\_cache\busts · high confidence
Security
Web chat UI gains host validation and atomic caching
The web chat UI now includes a \HostValidationMiddleware\ that rejects requests with unapproved \Host\ headers to prevent DNS rebinding attacks, configurable via the new \allowed\_hosts\ parameter in \Agent.to\_web()\ and \clai web\. Additionally, the bundled UI HTML is now fetched and written to the local cache atomically to prevent serving incomplete or corrupted files during crashes or concurrent writes.
_pydantic\_ai\_slim/pydantic\_ai/ui/\web · high confidence
Behavioural changes
10 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/macroscope
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/macroscope — 10 commits, 2 files: “Raise \UserError\ instead of \ModelRetry\ for \Macroscope\ setup failures (\#8999)”; “Give \AskUser\, \ConversationSearch\, \Macroscope\, \ManagedPrompt\, \PlaywrightBrowser\ and \PydanticAIDocs\ a stable default \id\, and append \ManagedPrompt\ version changes (\#10904)”; “Trim docs and deduplicate code in \pydantic-ai-harness\ integrations (\#12208)”; and 7 more (refs \#8999, \#10904, \#12208, \#10873, \#8830, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/macroscope · medium confidence · unverified
11 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/github
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/github — 11 commits, 2 files: “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Tighten the \GitHub\ harness docs' token-handling and connection prose (\#10860)”; “Trim docs and deduplicate code in \pydantic-ai-harness\ integrations (\#12208)”; and 8 more (refs \#8858, \#10860, \#12208, \#8830, \#11840, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/github · medium confidence · unverified
11 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/localstack
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/localstack — 11 commits, 3 files: “Tighten \LocalStack\ docs and instruction wording, and state its telemetry (\#10837)”; “Harden LocalStack capability after post-merge review (pydantic/pydantic-ai-harness\#425)”; “Deduplicate \pydantic-ai-harness\ sandbox provider checks and trim their docs (\#12205)”; and 8 more (refs \#10837, \#12205, \#8830, \#9854, \#11840, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/localstack · medium confidence · unverified
11 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/logfire
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/logfire — 11 commits, 2 files: “feat(logfire): add ManagedPrompt capability (pydantic/pydantic-ai-harness\#257)”; “docs: capability pages for the unified docs site + README/doc parity gate (pydantic/pydantic-ai-harness\#329)”; “Lead with the one-command Logfire wizard wherever the docs set up Logfire (\#11910)”; and 8 more (refs \#11910, \#10904, \#10867, \#10821, \#8830, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/logfire · medium confidence · unverified
11 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/pydantic\_ai\_docs
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/pydantic\_ai\_docs — 11 commits, 3 files: “Name models in the docs through tiered aliases defined in \docs/model\_aliases/aliases.json\ (\#10250)”; “Give \AskUser\, \ConversationSearch\, \Macroscope\, \ManagedPrompt\, \PlaywrightBrowser\ and \PydanticAIDocs\ a stable default \id\, and append \ManagedPrompt\ version changes (\#10904)”; “Trim docs and deduplicate code in \pydantic-ai-harness\ integrations (\#12208)”; and 8 more (refs \#10250, \#10904, \#12208, \#8829, \#9011, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/pydantic\_ai\docs · medium confidence · unverified
12 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/logfire\_mcp
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/logfire\mcp — 12 commits, 2 files: “Prefix \LogfireMCP\ tool names with \logfire\\ (\#9868)”; “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Trim docs and deduplicate code in \pydantic-ai-harness\ integrations (\#12208)”; and 9 more (refs \#9868, \#8858, \#12208, \#10543, \#10818, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/logfire\mcp · medium confidence · unverified
12 commits (0 fixes) modifying src/pydantic\_clai2/pydantic\_clai2/ui
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/ui — 12 commits, 5 files: “Log each \clai2\ Logfire session root's identity as soon as it opens, so running sessions can be queried (\#9867)”; “Keep a \$BROWSER\ opener's output from painting over the \clai2\ display (\#9995)”; “Keep macOS \MallocStackLogging\ noise from child processes out of the clai2 TUI (\#10063)”; and 9 more (refs \#9867, \#9995, \#10063, \#9797, \#9566, …).
_src/pydantic\_clai2/pydantic\clai2/ui · medium confidence · unverified
12 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/e2b\_sandbox
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/e2b\_sandbox — 12 commits (1 fix), 3 files: “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “Stop importing the undeclared \sniffio\ package in \pydantic-evals\ and \pydantic-ai-harness\ (\#9620)”; “Stop timed-out \E2BSandbox\ commands on templates without \ps\ (\#10809)”; and 9 more (refs \#11768, \#9620, \#10809, \#10811, \#12500, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/e2b\sandbox · medium confidence · unverified
129 commits (4 fixes) modifying src/pydantic\_clai2/pydantic\_clai2
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2 — 129 commits (4 fixs), 11 files: “Add \/effort\ to view and change reasoning effort in \clai2\ (\#9909)”; “Add opt-in \httpx\ instrumentation to the \clai2\ observability plugin (\#9921)”; “Select \CLAI2\ themes from Termflow with \/theme\ (pydantic/pydantic-ai-harness\#990)”; and 126 more (refs \#9909, \#9921, \#9932, \#9306, \#9808, …).
_src/pydantic\_clai2/pydantic\clai2 · medium confidence · unverified
13 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/advisor
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/advisor — 13 commits, 2 files: “docs: 'Your agent's favorite harness' tagline and an em-dash sweep (pydantic/pydantic-ai-harness\#627)”; “File internal agent runs under the parent run's conversation (pydantic/pydantic-ai-harness\#1069)”; “Trim docs, docstrings and duplicated helpers in \pydantic-ai-harness\ delegation and planning capabilities (\#12219)”; and 10 more (refs \#12219, \#10752, \#8858, \#8830, \#11840, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/advisor · medium confidence · unverified
13 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/background\_tools
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/background\_tools — 13 commits, 2 files: “Frame \BackgroundTools\ results as \\<tool\_result\>\ tool output instead of untagged user text (\#10548)”; “Report \ModelRetry\ and \ToolFailed\ from execute hooks on \BackgroundTools\ calls to the model (\#10830)”; “Keep finished \BackgroundTools\ results in the history when a run pauses for deferred tools (\#11752)”; and 10 more (refs \#10548, \#10830, \#11752, \#8858, \#10196, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/background\tools · medium confidence · unverified
13 commits (0 fixes) modifying src/pydantic\_clai2/pydantic\_clai2/builtin\_plugins
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/builtin\_plugins — 13 commits, 6 files: “Show the \coder\ plugin settings menu in the \clai2\ web UI (\#10334)”; “Keep a \$BROWSER\ opener's output from painting over the \clai2\ display (\#9995)”; “pydantic-clai2 0.61.0/new starts a session; /resume restores one; /exit quits. Esc or Ctrl-C interrupts a turn. customization guide and hosted integrations deferred capabilities (\#11337)”; and 10 more (refs \#10334, \#9995, \#11337, \#9598, \#9430, …).
_src/pydantic\_clai2/pydantic\_clai2/builtin\plugins · medium confidence · unverified
14 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/ask\_user
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/ask\_user — 14 commits, 3 files: “Let \AskUser\ defer questions to the host and time out a slow answerer (\#9509)”; “Allow custom answers to \ask\_user\_question\ in \CLAI\ (pydantic/pydantic-ai-harness\#1009)”; “Trim docs, docstrings and duplicated helpers in \pydantic-ai-harness\ delegation and planning capabilities (\#12219)”; and 11 more (refs \#9509, \#12219, \#11664, \#10904, \#10755, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/ask\user · medium confidence · unverified
14 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/trajectory\_judge
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/trajectory\_judge — 14 commits, 2 files: “File internal agent runs under the parent run's conversation (pydantic/pydantic-ai-harness\#1069)”; “Tie \TrajectoryJudge\'s steering note to its own frame with a \ref\, so forged \\<judge\_feedback\>\ tags get no authority (\#10869)”; “Point harness links and contribution guidance at pydantic-ai (\#8858)”; and 11 more (refs \#10869, \#8858, \#11530, \#10545, \#9012, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/trajectory\judge · medium confidence · unverified
14 commits (0 fixes) modifying src/pydantic\_clai2/pydantic\_clai2/models
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/models — 14 commits, 5 files: “Add \/effort\ to view and change reasoning effort in \clai2\ (\#9909)”; “Enable Anthropic prompt caching by default in \clai2\ (\#9859)”; “Manage \clai2\ fallback chains in the \/model\ picker, with \/chain\ as an alias of \/model chains\ (\#9838)”; and 11 more (refs \#9909, \#9859, \#9838, \#10920, \#9430, …).
_src/pydantic\_clai2/pydantic\clai2/models · medium confidence · unverified
14 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/aws\_lambda
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/aws\_lambda — 14 commits (1 fix), 3 files: “feat(aws-lambda): AWSLambdaDurability for AWS Lambda durable functions (pydantic/pydantic-ai-harness\#417)”; “Let a retired \aws\_lambda\ agent loop finish its own cleanup at the grace deadline (\#12140)”; “Point harness links and contribution guidance at pydantic-ai (\#8858)”; and 11 more (refs \#12140, \#8858, \#8832, \#8829, \#9654, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/aws\lambda · medium confidence · unverified
14 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/browser\_use
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/browser\_use — 14 commits (1 fix), 2 files: “Harden capability defaults and execution boundaries (pydantic/pydantic-ai-harness\#621)”; “feat(browseruse): delegate open-ended web tasks to a browser-use agent (pydantic/pydantic-ai-harness\#419)”; “Require Python 3.11 across the workspace (\#9526)”; and 11 more (refs \#9526, \#12208, \#10775, \#9506, \#8830, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/browser\use · medium confidence · unverified
149 commits (12 fixes) modifying pydantic\_ai\_slim/pydantic\_ai/models
A change to existing behaviour in pydantic\_ai\_slim/pydantic\_ai/models — 149 commits (12 fixs), 17 files: “Add \kimi-k3.1-pro\ and \kimi-k3.1-pro-highspeed\ model names (\#11173)”; “Replace HTTP client cache with \create\_async\_http\_client\ and context manager (\#4421)”; “Send stored response, item, conversation and container IDs back only to the endpoint that created them, via \Model.accepts\_server\_state\ (\#11774)”; and 146 more (refs \#11173, \#4421, \#11774, \#9350, \#11090, …).
_pydantic\_ai\_slim/pydantic\ai/models · medium confidence · unverified
15 commits (0 fixes) modifying pydantic\_ai\_slim/pydantic\_ai/models/anthropic
A change to existing behaviour in pydantic\_ai\_slim/pydantic\_ai/models/anthropic — 15 commits, 13 files: “Keep strict schemas within Anthropic's and OpenAI's complexity limits, falling back to non-strict output where no mode was chosen (\#11464)”; “Ignore \max\_retries\ in OpenAI, Anthropic and Gemini batch requests instead of raising \UserError\ (\#11931)”; “Call Anthropic's \20260209\ web tools directly on the Anthropic API when citations are requested (\#12278)”; and 12 more (refs \#11464, \#11931, \#12278, \#11934, \#11411, …).
_pydantic\_ai\_slim/pydantic\ai/models/anthropic · medium confidence · unverified
15 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/dynamic\_workflow
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/dynamic\_workflow — 15 commits (1 fix), 2 files: “docs: point at the new per-capability pages in core's Capabilities section (pydantic/pydantic-ai-harness\#391)”; “Bump Monty to 0.0.19 (pydantic/pydantic-ai-harness\#345)”; “docs: capability pages for the unified docs site + README/doc parity gate (pydantic/pydantic-ai-harness\#329)”; and 12 more (refs \#10497, \#11962, \#10672, \#8830, \#11840, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/dynamic\workflow · medium confidence · unverified
16 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/exa
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/exa — 16 commits (1 fix), 4 files: “Let capability specs set the \AbstractCapability\ base fields \id\, \description\ and \defer\_loading\ (\#11858)”; “feat: \web\_search\ \text\_summary\ mode and deferred \ExaAgent\ capability (pydantic/pydantic-ai-harness\#386)”; “Rebuild the default \ExaSearch\ and \ExaAgent\ client on a new event loop, and build \ExaAgent\'s once (\#10848)”; and 13 more (refs \#11858, \#10848, \#11097, \#10451, \#10840, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/exa · medium confidence · unverified
17 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/modal\_sandbox
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/modal\_sandbox — 17 commits, 3 files: “Document the \ModalSandbox\ network and \env\ exposure and tighten its docs (\#10819)”; “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “feat: add \ModalSandbox\ for isolated cloud sandboxes (pydantic/pydantic-ai-harness\#269)”; and 14 more (refs \#10819, \#11768, \#9620, \#10871, \#9654, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/modal\sandbox · medium confidence · unverified
17 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/skills
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/skills — 17 commits, 2 files: “Honour \disable-model-invocation\ and \user-invocable\ in \Skills\, skip stray skill library paths, and find \clai2\ repository skills from subdirectories (\#10378)”; “docs: 'Your agent's favorite harness' tagline and an em-dash sweep (pydantic/pydantic-ai-harness\#627)”; “Scope the \Skills\ id-collision note to skills exposed to the model (\#11428)”; and 14 more (refs \#10378, \#11428, \#12219, \#8858, \#11185, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/skills · medium confidence · unverified
17 commits (0 fixes) modifying src/pydantic\_clai2/pydantic\_clai2/mcp
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/mcp — 17 commits, 5 files: “Load a repository's Claude Code-style \.mcp.json\ in \clai2\ \/mcp\ (pydantic/pydantic-ai-harness\#1109)”; “Sign in to \clai2\ plugins and OAuth MCP servers only on request, through one shared sign-in API (\#9915)”; “Let a configured \clai2\ MCP server load its tools on demand with \defer\_loading\ (\#11629)”; and 14 more (refs \#9915, \#11629, \#8953, \#11656, \#8832, …).
_src/pydantic\_clai2/pydantic\clai2/mcp · medium confidence · unverified
18 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/system\_reminders
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/system\_reminders — 18 commits (1 fix), 2 files: “docs: 'Your agent's favorite harness' tagline and an em-dash sweep (pydantic/pydantic-ai-harness\#627)”; “Trim docs, docstrings and duplicated helpers in \pydantic-ai-harness\ context management capabilities (\#12224)”; “Show the harness \Planning\ plan in the plan tools' results instead of a per-request reminder copy (\#11521)”; and 15 more (refs \#12224, \#11521, \#8858, \#11471, \#11642, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/system\reminders · medium confidence · unverified
19 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/conversation\_search
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/conversation\_search — 19 commits, 4 files: “State when the \ConversationSearch\ default-scope warning will be removed (\#9383)”; “Harden capability defaults and execution boundaries (pydantic/pydantic-ai-harness\#621)”; “Add \RelevanceCompaction\, which keeps, reduces or drops each older history step by a judge model's ranking (\#11026)”; and 16 more (refs \#9383, \#11026, \#12224, \#8832, \#10904, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/conversation\search · medium confidence · unverified
2 commits (0 fixes) modifying .agents/skills/adopt-pr
A change to existing behaviour in .agents/skills/adopt-pr — 2 commits, 2 files: “Make contributor-agent PR workflows durable (\#7662)”; “Authorize takeover comments during \/adopt-pr\ (\#10915)” (refs \#7662, \#10915).
.agents/skills/adopt-pr · medium confidence · unverified
2 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/artifact\_scanning
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/artifact\_scanning — 2 commits, 2 files: “Strip every default-ignorable character before \ArtifactScanner\ phrase matching and flag hidden variation selectors (\#11475)”; “Add agent artifact security scanning for \Memory\ and \Skills\ with \ArtifactScanner\ (\#11470)” (refs \#11475, \#11470).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/artifact\scanning · medium confidence · unverified
23 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/repo\_context
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/repo\_context — 23 commits, 3 files: “Cap inlined nested instructions at 32,000 characters per conversation by default with \RepoContext(nested\_max\_chars=...)\ (\#11699)”; “State how nested instruction files relate in \RepoContext\ instructions when \nested\_traversal\ is on (\#11698)”; “Trim docs, docstrings and duplicated helpers in \pydantic-ai-harness\ context management capabilities (\#12224)”; and 20 more (refs \#11699, \#11698, \#12224, \#11691, \#11530, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/repo\context · medium confidence · unverified
23 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/planning
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/planning — 23 commits (1 fix), 5 files: “Emit \Planning\ status-tool events after their store reads, so DBOS recovery with an \event\_stream\_handler\ replays in order (\#10343)”; “Show the harness \Planning\ plan in the plan tools' results instead of a per-request reminder copy (\#11521)”; “Keep the registered \Planning\ toolset across runs so durable execution accepts it (\#9507)”; and 20 more (refs \#10343, \#11521, \#9507, \#12219, \#8837, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/planning · medium confidence · unverified
24 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/guardrails
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/guardrails — 24 commits, 5 files: “feat(guardrails): add \ToolGuard\ for tool arguments and tool results (pydantic/pydantic-ai-harness\#470)”; “Harden capability defaults and execution boundaries (pydantic/pydantic-ai-harness\#621)”; “docs: 'Your agent's favorite harness' tagline and an em-dash sweep (pydantic/pydantic-ai-harness\#627)”; and 21 more (refs \#12428, \#8858, \#10365, \#10841, \#10550, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/guardrails · medium confidence · unverified
24 commits (0 fixes) modifying src/pydantic\_clai2/pydantic\_clai2/plugins
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/plugins — 24 commits, 7 files: “Sign in to \clai2\ plugins and OAuth MCP servers only on request, through one shared sign-in API (\#9915)”; “Keep \compaction\ and a separate \SubAgents\ off in \clai2\ while \coder\ already provides them (\#9657)”; “Run several live sessions in one \clai2\ process with \LiveRuntimes\: \open\, \new\, \close\, and idle closing (\#10494)”; and 21 more (refs \#9915, \#9657, \#10494, \#11466, \#11423, …).
_src/pydantic\_clai2/pydantic\clai2/plugins · medium confidence · unverified
26 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/tool\_output\_limits
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/tool\_output\_limits — 26 commits, 3 files: “Count truncation markers within \Truncate.max\_chars\ (pydantic/pydantic-ai-harness\#816)”; “Trim docs, docstrings and duplicated helpers in \pydantic-ai-harness\ context management capabilities (\#12224)”; “File internal agent runs under the parent run's conversation (pydantic/pydantic-ai-harness\#1069)”; and 23 more (refs \#12224, \#10457, \#11453, \#8829, \#10339, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/tool\_output\limits · medium confidence · unverified
26 commits (2 fixes) modifying pydantic\_ai\_slim/pydantic\_ai/durable\_exec
A change to existing behaviour in pydantic\_ai\_slim/pydantic\_ai/durable\_exec — 26 commits (2 fixs), 7 files: “Give one-off capabilities a default \id\ and a \combine\ rule for repeats (\#7248)”; “Reject per-run \capabilities=\ on DBOS, matching Temporal (\#6955)”; “Bound a whole agent run with \timeout=\ and \RunContext.deadline\ (\#10070)”; and 23 more (refs \#7248, \#6955, \#10070, \#12345, \#8463, …).
_pydantic\_ai\_slim/pydantic\_ai/durable\exec · medium confidence · unverified
27 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/memory
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/memory — 27 commits (1 fix), 8 files: “Let capability specs set the \AbstractCapability\ base fields \id\, \description\ and \defer\_loading\ (\#11858)”; “docs: point at the new per-capability pages in core's Capabilities section (pydantic/pydantic-ai-harness\#391)”; “docs: 'Your agent's favorite harness' tagline and an em-dash sweep (pydantic/pydantic-ai-harness\#627)”; and 24 more (refs \#11858, \#9489, \#12224, \#8858, \#10892, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/memory · medium confidence · unverified
28 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/spend
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/spend — 28 commits (1 fix), 7 files: “feat(spend): add \SpendLimits\ for spend and token budgets across windows and processes (pydantic/pydantic-ai-harness\#474)”; “Make \SpendLimits\ replay-safe under durable execution (pydantic/pydantic-ai-harness\#748)”; “Charge \FallbackModel\ attempts once in \SpendLimits\ when a failed continuation chain is retried (\#10059)”; and 25 more (refs \#10059, \#9491, \#12219, \#9792, \#8858, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/spend · medium confidence · unverified
3 commits (0 fixes) modifying .agents/skills/branch-context
A change to existing behaviour in .agents/skills/branch-context — 3 commits, 10 files: “Make \append-handoff.sh --help\ print usage, and refuse a handoff without a lane (\#11050)”; “Make contributor-agent PR workflows durable (\#7662)”; “Keep confidential reasoning out of the public PR decisions section and record decision session ids (\#11566)” (refs \#11050, \#7662, \#11566).
.agents/skills/branch-context · medium confidence · unverified
3 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/detect\_loops
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/detect\_loops — 3 commits, 2 files: “Add a \DetectLoops\ capability to interrupt repeated-action loops (\#11471)”; “Add a diagram to every Pydantic AI Harness capability page (\#11840)”; “Pair \DetectLoops\ tool results with the response that made the calls (\#11551)” (refs \#11471, \#11840, \#11551).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/detect\loops · medium confidence · unverified
3 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/docs
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/docs — 3 commits, 1 file: “refactor: graduate capabilities out of \experimental\ (ACP excepted) (pydantic/pydantic-ai-harness\#347)”; “Keep pre-rename PyaiDocs agent specs loading (pydantic/pydantic-ai-harness\#488)”; “Rename capabilities to follow the documented naming convention (pydantic/pydantic-ai-harness\#480)”.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/docs · medium confidence · unverified
32 commits (3 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/filesystem
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/filesystem — 32 commits (3 fixs), 6 files: “Give \FileSystem\ a \max\_read\_chars\ budget so \Coder\ reads page without skipping lines (pydantic/pydantic-ai-harness\#892)”; “Announce file changes and report edits, directories, and searches as \file\_system\ events (pydantic/pydantic-ai-harness\#853)”; “Let \read\_file\ return images and PDFs, and describe other binary files (\#9596)”; and 29 more (refs \#9596, \#10674, \#11851, \#10825, \#10627, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/filesystem · medium confidence · unverified
33 commits (0 fixes) modifying src/pydantic\_clai2/pydantic\_clai2/ui/rendering
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/ui/rendering — 33 commits, 17 files: “Distinguish omitted provider usage from reported zero and add \UsageLimitUnavailableWarning\ (\#9491)”; “Name \ask\_user\_question\ calls in \clai2\ by question header instead of raw JSON (\#9907)”; “Import Claude Code sessions into \clai2\ with their parallel tool results, injected context, names, and sub-agents as tasks (\#11186)”; and 30 more (refs \#9491, \#9907, \#11186, \#9886, \#10469, …).
_src/pydantic\_clai2/pydantic\clai2/ui/rendering · medium confidence · unverified
34 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/subagents
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/subagents — 34 commits (1 fix), 4 files: “Steer a running delegated task with \DelegationTasks.steer\ (\#10359)”; “Support \SubAgents(dynamic\_agents=...)\ under durable execution (\#12418)”; “Trim docs, docstrings and duplicated helpers in \pydantic-ai-harness\ delegation and planning capabilities (\#12219)”; and 31 more (refs \#10359, \#12418, \#12219, \#11007, \#9023, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/subagents · medium confidence · unverified
36 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/coder
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/coder — 36 commits (1 fix), 5 files: “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “Expose \RepairToolArguments\ as a standalone capability (pydantic/pydantic-ai-harness\#917)”; “Cap inlined nested instructions at 32,000 characters per conversation by default with \RepoContext(nested\_max\_chars=...)\ (\#11699)”; and 33 more (refs \#11768, \#11699, \#9969, \#11691, \#12219, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/coder · medium confidence · unverified
36 commits (2 fixes) modifying .agents/skills/pushing-commits-to-the-repo
A change to existing behaviour in .agents/skills/pushing-commits-to-the-repo — 36 commits (2 fixs), 4 files: “Upgrade \douwebot\ to Opus 5.5 and submit review verdicts. (\#9680)”; “Refresh the GitHub label catalog (\#9785)”; “Attribute douwebot inline comments to the head they were made on in \PR Review Gates\ (\#12417)”; and 33 more (refs \#9680, \#9785, \#12417, \#9660, \#6795, …).
.agents/skills/pushing-commits-to-the-repo · medium confidence · unverified
37 commits (0 fixes) modifying src/pydantic\_clai2/pydantic\_clai2/cli
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/cli — 37 commits, 8 files: “Add \/effort\ to view and change reasoning effort in \clai2\ (\#9909)”; “Keep \compaction\ and a separate \SubAgents\ off in \clai2\ while \coder\ already provides them (\#9657)”; “Give \clai2\ plugins the conversation ID and title, and a \ConversationChanged\ event (\#9808)”; and 34 more (refs \#9909, \#9657, \#9808, \#10352, \#10494, …).
_src/pydantic\_clai2/pydantic\clai2/cli · medium confidence · unverified
37 commits (1 fix) modifying src/pydantic\_clai2/pydantic\_clai2/ui/prompt
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/ui/prompt — 37 commits (1 fix), 22 files: “Add a double-\Esc\ conversation rewind menu to \clai2\ (\#9688)”; “Stop an idle \clai2\ editor from rebuilding its frame ten times a second (\#10203)”; “Keep a \$BROWSER\ opener's output from painting over the \clai2\ display (\#9995)”; and 34 more (refs \#9688, \#10203, \#9995, \#9841, \#10477, …).
_src/pydantic\_clai2/pydantic\clai2/ui/prompt · medium confidence · unverified
4 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/boxd\_sandbox
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/boxd\_sandbox — 4 commits, 3 files: “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “Deduplicate \pydantic-ai-harness\ sandbox provider checks and trim their docs (\#12205)”; “Add \BoxdSandbox\: a workspace that runs \Coder\, \Shell\, \FileSystem\ and every other harness capability in a boxd machine (\#11367)”; and 1 more (refs \#11768, \#12205, \#11367, \#11840).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/boxd\sandbox · medium confidence · unverified
4 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/experimental/docs
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/experimental/docs — 4 commits, 1 file: “Add experimental CE-discovery capabilities: context, subagents disk-load, docs, authoring (pydantic/pydantic-ai-harness\#304)”; “refactor: graduate capabilities out of \experimental\ (ACP excepted) (pydantic/pydantic-ai-harness\#347)”; “Keep pre-rename PyaiDocs agent specs loading (pydantic/pydantic-ai-harness\#488)”; and 1 more.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/experimental/docs · medium confidence · unverified
4 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/pgtask
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/pgtask — 4 commits, 3 files: “Add \PGTaskDurability\ to the harness as a replacement for \pydantic-ai-pgtask\ (\#10174)”; “Add provider batch API support with a \BatchModelRequests\ capability and a \get\_wrapper\_model\ capability hook (\#10672)”; “Add a diagram to every Pydantic AI Harness capability page (\#11840)”; and 1 more (refs \#10174, \#10672, \#11840, \#11163).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/pgtask · medium confidence · unverified
4 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/smart\_search
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/smart\_search — 4 commits, 4 files: “Trim docs, docstrings and duplicated helpers in \pydantic-ai-harness\ context management capabilities (\#12224)”; “Add a diagram to every Pydantic AI Harness capability page (\#11840)”; “Rename the \SmartFileSearch\ harness capability to \SmartSearch\ (\#11016)”; and 1 more (refs \#12224, \#11840, \#11016, \#12415).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/smart\search · medium confidence · unverified
4 commits (0 fixes) modifying src/pydantic\_clai2/scripts
A change to existing behaviour in src/pydantic\_clai2/scripts — 4 commits, 23 files: “clai2: publish bleeding sdists to a \clai2-bleeding\ release and install \/update\ from it (\#9872)”; “Rewrite the \pydantic-clai2\ README as a product page with screenshots, moving the reference to \docs/reference.md\ (\#10406)”; “Rename the \clai2\ \updates.channel\ value \bleeding\ to \main\ (\#9892)”; and 1 more (refs \#9872, \#10406, \#9892, \#11043).
_src/pydantic\clai2/scripts · medium confidence · unverified
40 commits (0 fixes) modifying src/pydantic\_clai2/pydantic\_clai2/runtime
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/runtime — 40 commits, 9 files: “Steer a running delegated task with \DelegationTasks.steer\ (\#10359)”; “Give \clai2\ plugins the conversation ID and title, and a \ConversationChanged\ event (\#9808)”; “Run each \clai2 --web\ session in its own directory, and ask where a new one runs (\#10352)”; and 37 more (refs \#10359, \#9808, \#10352, \#10494, \#9499, …).
_src/pydantic\_clai2/pydantic\clai2/runtime · medium confidence · unverified
46 commits (2 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/step\_persistence
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/step\_persistence — 46 commits (2 fixs), 9 files: “Close \SqliteStepStore\ and \SqliteMediaStore\ connections whose WAL switch fails on a locked database (\#12147)”; “Document when an interrupted \StepPersistence\ snapshot needs \repair\_messages\ (\#9602)”; “docs: 'Your agent's favorite harness' tagline and an em-dash sweep (pydantic/pydantic-ai-harness\#627)”; and 43 more (refs \#12147, \#9602, \#9499, \#12219, \#11614, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/step\persistence · medium confidence · unverified
47 commits (0 fixes) modifying src/pydantic\_clai2/pydantic\_clai2/config
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/config — 47 commits, 10 files: “Add opt-in \httpx\ instrumentation to the \clai2\ observability plugin (\#9921)”; “Honour \disable-model-invocation\ and \user-invocable\ in \Skills\, skip stray skill library paths, and find \clai2\ repository skills from subdirectories (\#10378)”; “CLAI2: \/update\ with stable (PyPI) and bleeding (\main\) channels (\#9576)”; and 44 more (refs \#9921, \#10378, \#9576, \#11656, \#10375, …).
_src/pydantic\_clai2/pydantic\clai2/config · medium confidence · unverified
49 commits (1 fix) modifying pydantic\_ai\_slim/pydantic\_ai/realtime
A change to existing behaviour in pydantic\_ai\_slim/pydantic\_ai/realtime — 49 commits (1 fix), 9 files: “Raise a \UserError\ for a realtime \tool\_choice\ that forces a tool call on OpenAI, Azure OpenAI, and xAI (\#8755)”; “Record what the user heard of ElevenLabs Agents replies they barged in on, adding the \OutputItemCut\ lifecycle event (\#10739)”; “Check \include\_content=False\ over every content channel at once (\#8419)”; and 46 more (refs \#8755, \#10739, \#8419, \#8132, \#11685, …).
_pydantic\_ai\_slim/pydantic\ai/realtime · medium confidence · unverified
49 commits (5 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/code\_mode
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/code\_mode — 49 commits (5 fixs), 6 files: “Rename package to pydantic-ai-harness (pydantic/pydantic-ai-harness\#207)”; “Harden capability defaults and execution boundaries (pydantic/pydantic-ai-harness\#621)”; “fix(code\_mode): tighten \run\_code\ prompt against phantom imports and bad inputs (pydantic/pydantic-ai-harness\#295)”; and 46 more (refs \#11141, \#11358, \#10137, \#10365, \#9710, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/code\mode · medium confidence · unverified
5 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/render
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/render — 5 commits, 3 files: “Add provider batch API support with a \BatchModelRequests\ capability and a \get\_wrapper\_model\ capability hook (\#10672)”; “Add a diagram to every Pydantic AI Harness capability page (\#11840)”; “Adds Render Workflows Integration for Durable Execution engines (\#9439)”; and 2 more (refs \#10672, \#11840, \#9439, \#12216, \#10238).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/render · medium confidence · unverified
5 commits (0 fixes) modifying src/pydantic\_clai2/web-ui
A change to existing behaviour in src/pydantic\_clai2/web-ui — 5 commits, 5 files: “Show the Pydantic logo in the \clai2\ web UI (\#10274)”; “Use Pydantic brand colours in the \clai2\ web UI, in dark and light mode (\#10269)”; “Stop committing the built \clai2\ web UI bundle (\app.js\) in pull requests, so they stop getting ejected from the merge queue (\#11563)”; and 2 more (refs \#10274, \#10269, \#11563, \#10263, \#11043).
_src/pydantic\clai2/web-ui · medium confidence · unverified
52 commits (5 fixes) modifying scripts
A change to existing behaviour in scripts — 52 commits (5 fixs), 14 files: “Send stored response, item, conversation and container IDs back only to the endpoint that created them, via \Model.accepts\_server\_state\ (\#11774)”; “Add Azure Durable Task durability to \pydantic\_ai\_harness\ (\#11332)”; “Stop Temporal from retrying model errors that can't succeed on retry by default, via \non\_retryable\_model\_errors\ (\#12107)”; and 49 more (refs \#11774, \#11332, \#12107, \#4291, \#8893, …).
scripts · medium confidence · unverified
6 commits (0 fixes) modifying pydantic\_ai\_slim/pydantic\_ai/workspaces
A change to existing behaviour in pydantic\_ai\_slim/pydantic\_ai/workspaces — 6 commits, 7 files: “Fail closed in the shell \realpath\ fallback when \readlink\ cannot read a symlink (\#9613)”; “Add per-path workspace authority with \Workspace.authority\_of\ (\#12473)”; “Keep planted files from escaping \BubblewrapSandbox\ on the next launch (\#9457)”; and 3 more (refs \#9613, \#12473, \#9457, \#9854, \#11836, …).
_pydantic\_ai\_slim/pydantic\ai/workspaces · medium confidence · unverified
6 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/absurd
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/absurd — 6 commits, 4 files: “Keep streamed model deltas out of durable execution histories by replaying the final response (\#10958)”; “Add \AbsurdDurability\ to the harness as a replacement for \pydantic-ai-absurd\ (\#8946)”; “Point \pydantic-ai-harness\ and \pydantic-clai2\ at \pydantic/pydantic-ai\ instead of the old \pydantic/pydantic-ai-harness\ repo (\#9651)”; and 3 more (refs \#10958, \#8946, \#9651, \#10672, \#11840, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/absurd · medium confidence · unverified
6 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/repair\_tool\_arguments
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/repair\_tool\_arguments — 6 commits, 3 files: “Write \RepairToolArguments\ repairs back to the \ToolCallPart\ in history (\#12319)”; “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “Expose \RepairToolArguments\ as a standalone capability (pydantic/pydantic-ai-harness\#917)”; and 3 more (refs \#12319, \#11768, \#8829, \#8830, \#11840).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/repair\_tool\arguments · medium confidence · unverified
6 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/smol\_sandbox
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/smol\_sandbox — 6 commits, 6 files: “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “Add BoxLiteSandbox: a local BoxLite micro-VM workspace (\#10740)”; “Deduplicate \pydantic-ai-harness\ sandbox provider checks and trim their docs (\#12205)”; and 3 more (refs \#11768, \#10740, \#12205, \#11840, \#10882, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/smol\sandbox · medium confidence · unverified
62 commits (0 fixes) modifying src/pydantic\_clai2/pydantic\_clai2/builtin\_plugins/web/static
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/builtin\_plugins/web/static — 62 commits, 3 files: “Add plugin side panels to the \clai2\ web UI, with a Logfire live view panel (\#10357)”; “Steer a running delegated task with \DelegationTasks.steer\ (\#10359)”; “Add the web UI's Orchestrator chat, a separate agent that runs \workspaces\ (\#10374)”; and 59 more (refs \#10357, \#10359, \#10374, \#10445, \#12008, …).
_src/pydantic\_clai2/pydantic\_clai2/builtin\plugins/web/static · medium confidence · unverified
62 commits (1 fix) modifying src/pydantic\_clai2/pydantic\_clai2/ui/menus
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/ui/menus — 62 commits (1 fix), 23 files: “Add a double-\Esc\ conversation rewind menu to \clai2\ (\#9688)”; “Keep \compaction\ and a separate \SubAgents\ off in \clai2\ while \coder\ already provides them (\#9657)”; “Give full-screen \clai2\ \!\ commands such as \!nano\ the terminal instead of capturing their output (\#10188)”; and 59 more (refs \#9688, \#9657, \#10188, \#9576, \#10461, …).
_src/pydantic\_clai2/pydantic\clai2/ui/menus · medium confidence · unverified
63 commits (3 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/compaction
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/compaction — 63 commits (3 fixs), 6 files: “Keep tool search discoveries across \SummarizingCompaction\, \SlidingWindowCompaction\ and \RelevanceCompaction\ by recording them again as a \ToolAvailabilityDeltaPart\ (\#11773)”; “Escape framing tags consistently in the \SummarizingCompaction\ summarizer transcript (\#11583)”; “Add \FallbackCompaction\ for failure-based strategy chains (pydantic/pydantic-ai-harness\#644)”; and 60 more (refs \#11773, \#11583, \#11026, \#12224, \#11521, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/compaction · medium confidence · unverified
66 commits (0 fixes) modifying src/pydantic\_clai2/web-ui/src/components
A change to existing behaviour in src/pydantic\_clai2/web-ui/src/components — 66 commits, 33 files: “Show a running shell command's live output in its \clai2\ web UI card (\#12221)”; “Validate the \clai2\ web sidebar's saved state and align its docs with the Archived view (\#11705)”; “Add plugin side panels to the \clai2\ web UI, with a Logfire live view panel (\#10357)”; and 63 more (refs \#12221, \#11705, \#10357, \#10359, \#10374, …).
_src/pydantic\clai2/web-ui/src/components · medium confidence · unverified
66 commits (7 fixes) modifying pydantic\_ai\_slim/pydantic\_ai/ui/ag\_ui
A change to existing behaviour in pydantic\_ai\_slim/pydantic\_ai/ui/ag\_ui — 66 commits (7 fixs), 8 files: “Stream \ctx.enqueue\ messages to Vercel AI and AG-UI frontends as history they send back but don't render (\#10405)”; “feat: add \conversation\_id\ for cross-run correlation (\#5251)”; “Preserve Anthropic recovery across normalized history (\#8040)”; and 63 more (refs \#10405, \#5251, \#8040, \#10563, \#5866, …).
_pydantic\_ai\_slim/pydantic\_ai/ui/ag\ui · medium confidence · unverified
69 commits (0 fixes) modifying src/pydantic\_clai2/pydantic\_clai2/builtin\_plugins/web
A change to existing behaviour in src/pydantic\_clai2/pydantic\_clai2/builtin\_plugins/web — 69 commits, 23 files: “Add plugin side panels to the \clai2\ web UI, with a Logfire live view panel (\#10357)”; “Steer a running delegated task with \DelegationTasks.steer\ (\#10359)”; “Add the web UI's Orchestrator chat, a separate agent that runs \workspaces\ (\#10374)”; and 66 more (refs \#10357, \#10359, \#10374, \#9491, \#10352, …).
_src/pydantic\_clai2/pydantic\_clai2/builtin\plugins/web · medium confidence · unverified
7 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/daytona\_sandbox
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/daytona\_sandbox — 7 commits, 3 files: “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “Remove the Daytona \close()\ workaround and require \daytona\>=0.226.1\ (\#11977)”; “Deduplicate \pydantic-ai-harness\ sandbox provider checks and trim their docs (\#12205)”; and 4 more (refs \#11768, \#11977, \#12205, \#11840, \#12509, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/daytona\sandbox · medium confidence · unverified
7 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/ordinal
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/ordinal — 7 commits, 2 files: “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Trim docs and deduplicate code in \pydantic-ai-harness\ integrations (\#12208)”; “Point the harness docs at the monorepo (\#8830)”; and 4 more (refs \#8858, \#12208, \#8830, \#11840, \#10376, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/ordinal · medium confidence · unverified
7 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/tool\_call\_judge
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/tool\_call\_judge — 7 commits, 4 files: “Document using \ApprovalReviewer\ to answer \ToolCallJudge\ escalations (\#10772)”; “Show \ApprovalReviewer\ and \ToolCallJudge\ the arguments as written, and correct the approval reviewer docs (\#10727)”; “Add Harness \ApprovalReviewer\ to let a model approve low-risk tool calls before a person sees them (\#10143)”; and 4 more (refs \#10772, \#10727, \#10143, \#9041, \#12207, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/tool\_call\judge · medium confidence · unverified
7 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/sprites\_sandbox
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/sprites\_sandbox — 7 commits (1 fix), 2 files: “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “Raise \WorkspaceUnavailableError\ when a Sprite is deleted during a command, whatever its exit code (\#10193)”; “Add \SpritesSandbox\: a workspace that runs \Coder\, \Shell\, \FileSystem\ and every other harness capability in a Fly.io Sprite (\#8869)”; and 4 more (refs \#11768, \#10193, \#8869, \#12205, \#11840, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/sprites\sandbox · medium confidence · unverified
70 commits (0 fixes) modifying src/pydantic\_clai2/web-ui/src
A change to existing behaviour in src/pydantic\_clai2/web-ui/src — 70 commits, 19 files: “Validate the \clai2\ web sidebar's saved state and align its docs with the Archived view (\#11705)”; “Add plugin side panels to the \clai2\ web UI, with a Logfire live view panel (\#10357)”; “Steer a running delegated task with \DelegationTasks.steer\ (\#10359)”; and 67 more (refs \#11705, \#10357, \#10359, \#11984, \#10374, …).
_src/pydantic\clai2/web-ui/src · medium confidence · unverified
8 commits (0 fixes) modifying src/pydantic\_ai\_harness/examples
A change to existing behaviour in src/pydantic\_ai\_harness/examples — 8 commits, 4 files: “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “Lint \pydantic-ai-harness\ and \pydantic-clai2\ with the root Ruff config (\#8829)”; “Add \Pixeltable\ tools and memory storage to Harness (\#9369)”; and 5 more (refs \#11768, \#8829, \#9369, \#8830, \#12216, …).
_src/pydantic\_ai\harness/examples · medium confidence · unverified
8 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/day\_ai
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/day\_ai — 8 commits, 2 files: “Treat \auth=''\ as unset alongside \client\ in \DayAI\ and warn about tokens in Temporal history (\#10718)”; “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Trim docs and deduplicate code in \pydantic-ai-harness\ integrations (\#12208)”; and 5 more (refs \#10718, \#8858, \#12208, \#8830, \#11840, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/day\ai · medium confidence · unverified
8 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/docker\_sandbox
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/docker\_sandbox — 8 commits, 3 files: “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “Document running \DockerSandbox\ against a remote Docker host (\#11018)”; “Run the \pydantic-ai-harness\ suite under the blocking-call detector (\#11872)”; and 5 more (refs \#11768, \#11018, \#11872, \#9341, \#10996, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/docker\sandbox · medium confidence · unverified
8 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/notion
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/notion — 8 commits, 2 files: “Add hosted MCP tools with \Notion\ (pydantic/pydantic-ai-harness\#797)”; “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Trim docs and deduplicate code in \pydantic-ai-harness\ integrations (\#12208)”; and 5 more (refs \#8858, \#12208, \#8830, \#11840, \#10376, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/notion · medium confidence · unverified
8 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/playwright
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/playwright — 8 commits, 2 files: “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Give \AskUser\, \ConversationSearch\, \Macroscope\, \ManagedPrompt\, \PlaywrightBrowser\ and \PydanticAIDocs\ a stable default \id\, and append \ManagedPrompt\ version changes (\#10904)”; “Add a \PlaywrightBrowser\ capability driving Chromium through async Playwright (pydantic/pydantic-ai-harness\#420)”; and 5 more (refs \#8858, \#10904, \#12380, \#8830, \#11840).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/playwright · medium confidence · unverified
8 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/posthog
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/posthog — 8 commits, 2 files: “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Trim docs and deduplicate code in \pydantic-ai-harness\ integrations (\#12208)”; “Point the harness docs at the monorepo (\#8830)”; and 5 more (refs \#8858, \#12208, \#8830, \#11840, \#10376, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/posthog · medium confidence · unverified
8 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/pylon
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/pylon — 8 commits, 2 files: “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Trim docs and deduplicate code in \pydantic-ai-harness\ integrations (\#12208)”; “Point the harness docs at the monorepo (\#8830)”; and 5 more (refs \#8858, \#12208, \#8830, \#11840, \#10376, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/pylon · medium confidence · unverified
8 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/slack
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/slack — 8 commits, 2 files: “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Trim docs and deduplicate code in \pydantic-ai-harness\ integrations (\#12208)”; “Point the harness docs at the monorepo (\#8830)”; and 5 more (refs \#8858, \#12208, \#8830, \#11840, \#10376, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/slack · medium confidence · unverified
8 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/ssh\_workspace
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/ssh\_workspace — 8 commits, 3 files: “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “Add \SSHWorkspace\ and \BubblewrapSandbox\ harness capabilities (\#8956)”; “Keep planted files from escaping \BubblewrapSandbox\ on the next launch (\#9457)”; and 5 more (refs \#11768, \#8956, \#9457, \#11479, \#11004, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/ssh\workspace · medium confidence · unverified
8 commits (1 fix) modifying pydantic\_ai\_slim/pydantic\_ai/\_cli
A change to existing behaviour in pydantic\_ai\_slim/pydantic\_ai/\_cli — 8 commits (1 fix), 1 file: “Add \openai-chat:\ prefix and warn on bare \openai:\ which will switch to use the Responses API in v2 (\#5334)”; “Add Web Chat UI for any agent that can be launched using \clai web\ or \Agent.to\_web()\ (\#3456)”; “Rename built-in tools → native tools; register via \capabilities=\[NativeTool(...)\]\ (\#5338)”; and 5 more (refs \#5334, \#3456, \#5338, \#7437, \#4345, …).
_pydantic\_ai\_slim/pydantic\_ai/\cli · medium confidence · unverified
85 commits (2 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness — 85 commits (2 fixs), 36 files: “Rename package to pydantic-ai-harness (pydantic/pydantic-ai-harness\#207)”; “Harden capability defaults and execution boundaries (pydantic/pydantic-ai-harness\#621)”; “Add \YouSearch\ and \YouResearch\ capabilities backed by the You.com APIs (pydantic/pydantic-ai-harness\#646)”; and 82 more (refs \#8956, \#11026, \#9620, \#10548, \#11332, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness · medium confidence · unverified
9 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/google\_workspace
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/google\_workspace — 9 commits, 2 files: “Add hosted MCP tools with \GoogleWorkspace\ (pydantic/pydantic-ai-harness\#791)”; “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Add \google\_workspace\ as a built-in \clai2\ plugin backed by harness \GoogleWorkspace\ (\#8907)”; and 6 more (refs \#8858, \#8907, \#12208, \#8830, \#10800, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/google\workspace · medium confidence · unverified
9 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/grain
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/grain — 9 commits, 2 files: “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Trim docs and deduplicate code in \pydantic-ai-harness\ integrations (\#12208)”; “Point the harness docs at the monorepo (\#8830)”; and 6 more (refs \#8858, \#12208, \#8830, \#11840, \#10376, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/grain · medium confidence · unverified
9 commits (0 fixes) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/linear
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/linear — 9 commits, 2 files: “Point harness links and contribution guidance at pydantic-ai (\#8858)”; “Keep \Linear(id=None)\ uncombined like \GitHub(id=None)\ (\#10813)”; “Add hosted MCP tools with \Linear\ (pydantic/pydantic-ai-harness\#793)”; and 6 more (refs \#8858, \#10813, \#8830, \#11840, \#10376, …).
_src/pydantic\_ai\_harness/pydantic\_ai\harness/linear · medium confidence · unverified
9 commits (1 fix) modifying src/pydantic\_ai\_harness/pydantic\_ai\_harness/bubblewrap\_sandbox
A change to existing behaviour in src/pydantic\_ai\_harness/pydantic\_ai\_harness/bubblewrap\_sandbox — 9 commits (1 fix), 4 files: “Make \json-repair\ an optional dependency, needed only by \Coder\ and \RepairToolArguments\ (\#11768)”; “Add \SSHWorkspace\ and \BubblewrapSandbox\ harness capabilities (\#8956)”; “Keep planted files from escaping \BubblewrapSandbox\ on the next launch (\#9457)”; and 6 more (refs \#11768, \#8956, \#9457, \#9654, \#9349, …).
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/bubblewrap\sandbox · medium confidence · unverified
Compaction module graduated from experimental with backward-compatible imports
The compaction capability has moved out of the experimental namespace to \pydantic\_ai\_harness.compaction\. The previous location (\pydantic\_ai\_harness.experimental.compaction\) remains available but now emits a deprecation warning, ensuring existing code continues to work while encouraging migration to the stable import path. The module exposes compaction strategies such as \ClampOversizedMessages\, \SlidingWindowCompaction\, and \SummarizingCompaction\, along with utility functions for token estimation and message pinning.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/experimental/compaction · high confidence
Deprecate experimental authoring and context import paths
The \pydantic\_ai\_harness.experimental.authoring\ and \pydantic\_ai\_harness.experimental.context\ modules have been moved to \pydantic\_ai\_harness.capability\_creation\ and \pydantic\_ai\_harness.repo\_context\ respectively. Importing from the old experimental locations now triggers a \DeprecationWarning\, signaling that users should update their imports to the new stable paths to avoid future breakage.
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/experimental/authoring, src/pydantic\_ai\_harness/pydantic\_ai\harness/experimental/context · high confidence
Deprecated import paths for renamed capabilities
Several internal modules have been renamed to follow a consistent naming convention, and the old import locations are now deprecated. Importing from \pydantic\_ai\_harness.context\ now emits a warning and redirects to \pydantic\_ai\_harness.repo\_context\. Similarly, \pydantic\_ai\_harness.overflowing\_tool\_output\ redirects to \pydantic\_ai\_harness.tool\_output\_limits\, and \pydantic\_ai\_harness.runtime\_authoring\ redirects to \pydantic\_ai\_harness.capability\_creation\. Users should update their imports to use the new module names to avoid deprecation warnings.
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/context, src/pydantic\_ai\_harness/pydantic\_ai\_harness/overflowing\_tool\_output, src/pydantic\_ai\_harness/pydantic\_ai\_harness/runtime\authoring · high confidence
Deprecation of cache\_stability module in favor of warn\_on\_cache\_busts
The \cache\_stability\ module has been deprecated and renamed to \warn\_on\_cache\_busts\. Importing from the old location now emits a \DeprecationWarning\ and redirects to the new module, ensuring backward compatibility while encouraging users to update their imports to \pydantic\_ai\_harness.warn\_on\_cache\_busts\.
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/cache\stability · high confidence
Deprecation warnings for experimental media and step\_persistence imports
Importing from the \experimental.media\ and \experimental.step\_persistence\ modules now emits a \DeprecationWarning\. These capabilities have graduated out of the experimental namespace, and users should update their imports to use \pydantic\_ai\_harness.media\ and \pydantic\_ai\_harness.step\_persistence\ respectively to avoid these warnings.
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/experimental/media, src/pydantic\_ai\_harness/pydantic\_ai\_harness/experimental/step\persistence · high confidence
Dynamic workflow module moved out of experimental
The dynamic workflow capability has graduated from the experimental namespace to the stable \pydantic\_ai\_harness.dynamic\_workflow\ module. Importing from the old \pydantic\_ai\_harness.experimental.dynamic\_workflow\ path still works but now emits a DeprecationWarning, so users should update their imports to use the new stable location.
_src/pydantic\_ai\_harness/pydantic\_ai\_harness/experimental/dynamic\workflow · high confidence
Empty common\_tools module added
An empty \_\init\\_.py file was added to the pydantic\_ai\_slim/pydantic\_ai/common\_tools directory. While the commit message indicates the addition of a DuckDuckGoSearch tool, the provided diff shows no actual tool implementation or code changes, only the creation of an empty module file.
_pydantic\_ai\_slim/pydantic\_ai/common\tools · low confidence
Introduce ModelProfile to centralize model-family capability configuration
The library now uses a \ModelProfile\ system to define model-family capability facts and schema/request quirks independently of provider classes. This new architecture centralizes configuration for features such as structured output, native tool support, thinking/reasoning behavior, and JSON schema transformation. Provider-specific profiles (e.g., Anthropic, Google, Grok, OpenAI) now explicitly declare these capabilities, allowing the framework to handle model-specific behaviors like thinking level snapping, forced tool choice restrictions, and cache retention rules in a unified, maintainable way.
_pydantic\_ai\_slim/pydantic\ai/profiles · high confidence
Introduce dedicated number and duration formatting for evaluation reports
The evaluation reporting system now uses a new \render\_numbers\ module to format numerical metrics and durations. This change ensures that non-finite values (infinity and NaN) are explicitly rendered as \inf\, \-inf\, or \nan\ instead of causing errors, and applies consistent significant-figure rules to floats and integers. Duration metrics are now formatted with appropriate precision (microseconds, milliseconds, or seconds) based on their magnitude, and diff calculations include signed absolute differences with relative percentage or multiplier indicators where meaningful.
_pydantic\_evals/pydantic\evals/reporting · high confidence
Overflow capability moved to tool\_output\_limits with deprecation warning
The overflow capability has graduated out of the experimental directory and is now located in \pydantic\_ai\_harness.tool\_output\_limits\. Importing from the old \pydantic\_ai\_harness.experimental.overflow\ path still works but emits a \DeprecationWarning\, so users should update their imports to use the new \tool\_output\_limits\ module to avoid warnings and ensure future compatibility.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/experimental/overflow · high confidence
Planning capability moved out of experimental
The planning capability has graduated from the experimental phase. While importing from the old experimental path still works, it now emits a deprecation warning; users should update their imports to use \pydantic\_ai\_harness.planning\ instead.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/experimental/planning · high confidence
Pydantic AI v2: Major API overhaul with native tools, capabilities, and streaming events
This release introduces Pydantic AI v2, a significant update that replaces the previous API with a new architecture centered on capabilities and native tools. Users will now interact with \Agent\ instances that support a \capabilities\ list for feature detection, and built-in tools (such as \WebFetchTool\, \FileSearchTool\, and \MCPServerTool\) are registered via \NativeTool\ definitions rather than legacy function wrappers. The run lifecycle is now driven by a rich event stream (\AgentRunEvents\), exposing granular events like \ToolCallPart\, \ThinkingPart\, and \OutputToolCallEvent\ for better observability and control. Additionally, the library has migrated to \httpx2\ for HTTP clients, added support for streaming tool calls and reasoning content, and introduced new output modes like \NativeOutput\ and \PromptedOutput\. The CLI has also been renamed to \clai\.
_pydantic\_ai\_slim/pydantic\ai · high confidence
Removal of legacy agent, function call, and result modules
The \pydantic\_ai\ package has removed the \agent.py\, \function\_calls.py\, and \result.py\ modules. This eliminates the previous \Agent\ class implementation, the \CallInfo\ and system prompt protocols, and the \RunResult\/\RunStreamResult\ data structures, indicating a significant structural change to how agents are defined and how responses are handled.
_pydantic\ai · high confidence
Repository migration to uv workspace and updated developer experience
The project has migrated its dependency management from rye to uv, replacing the old \requirements.lock\ and \requirements-dev.lock\ files with a \uv.lock\ workspace file. This change is accompanied by a comprehensive overhaul of the developer workflow: the \Makefile\ now uses \uv\ for all tasks (install, lint, format, typecheck, test), pre-commit hooks have been reconfigured to use \uv run\ for tools like ruff and pyright, and the \.gitignore\ has been expanded to exclude new local state directories (e.g., \.vscode/\, \local-notes/\) and agent skill artifacts. Additionally, the repository now includes \AGENTS.md\ to guide AI contributors and \CLAUDE.md\ to point to it, standardizing how coding agents interact with the codebase.
(repo-wide) · high confidence
SubAgents module graduated from experimental to stable
The SubAgents capability has moved out of the experimental directory and is now available directly from \pydantic\_ai\_harness.subagents\. The previous import path \pydantic\_ai\_harness.experimental.subagents\ remains functional but now emits a \DeprecationWarning\, encouraging users to update their imports to the new stable location.
_src/pydantic\_ai\_harness/pydantic\_ai\harness/experimental/subagents · high confidence
pydantic-ai: BehavioralChange
The pydantic-ai library has undergone a major v2 preparation release, introducing several breaking changes and behavioral updates. The most significant change is the renaming of the \result\ attribute to \output\ on agent run results, affecting how users access the final output of an agent run. Additionally, the \result\ method on \StreamedRunResult\ has been removed in favor of property-style accessors like \.usage\, \.timestamp\, and \.response\. The library has also dropped support for Python 3.9, requiring Python 3.10 or later. Other behavioral changes include the deprecation of specifying a model name without a provider prefix, the removal of the Python evaluator for security reasons, and changes to how evaluation reports and report cases are structured as generic dataclasses. The release also includes updates to the web chat UI, support for new models across various providers, and improvements to streaming and tool handling. Also in this area: Introduce native realtime speech-to-speech session support; CLAI 2 developer documentation and plugin architecture guide; Improved CLAI2 session resumption, UI stability, and evaluation performance; Python 3.11 is now the minimum supported version; Prompt-cache health reporting, OpenAI instruction caching, and WebRTC hang-up support; and 16 more.
pydantic-ai · high confidence
Test coverage
1 commit adding/updating tests in src/pydantic\_ai\_harness/integration\_tests/azure\_durable\_task; 10 commits adding/updating tests in tests/harness/linear; 10 commits adding/updating tests in tests/harness/render; 11 commits adding/updating tests in tests/harness/github; 12 commits adding/updating tests in tests/harness/ask\_user; 120 commits adding/updating tests in tests/profiles; 14 commits adding/updating tests in tests/harness/e2b\_sandbox; 14 commits adding/updating tests in tests/harness/exa; 14 commits adding/updating tests in tests/harness/trajectory\_judge; 15 commits adding/updating tests in tests/harness/playwright; 16 commits adding/updating tests in tests/harness/browser\_use; 17 commits adding/updating tests in tests/harness/advisor; 17 commits adding/updating tests in tests/harness/skills; 18 commits adding/updating tests in tests/harness/aws\_lambda; 18 commits adding/updating tests in tests/models/openrouter; 183 commits adding/updating tests in tests/harness; 19 commits adding/updating tests in tests/harness/system\_reminders; 2 commits adding/updating tests in tests/harness/approval\_reviewer; 2 commits adding/updating tests in tests/harness/artifact\_scanning; 2 commits adding/updating tests in tests/harness/azure\_durable\_task; 2 commits adding/updating tests in tests/harness/render\_sandbox; 2 commits adding/updating tests in tests/harness/smol\_sandbox; 22 commits adding/updating tests in tests/models/bedrock; 24 commits adding/updating tests in tests/harness/repo\_context; 25 commits adding/updating tests in tests/harness/memory; 27 commits adding/updating tests in tests/harness/background\_tools; 29 commits adding/updating tests in tests/harness/guardrails; 3 commits adding/updating tests in tests/harness/boxd\_sandbox; 3 commits adding/updating tests in tests/harness/detect\_loops; 3 commits adding/updating tests in tests/harness/docker\_sandbox; 3 commits adding/updating tests in tests/harness/pgtask; 37 commits adding/updating tests in tests/models/google; 4 commits adding/updating tests in tests/harness/absurd; 4 commits adding/updating tests in tests/harness/boxlite\_sandbox; 4 commits adding/updating tests in tests/harness/ipython\_code\_mode; 41 commits adding/updating tests in tests/harness/step\_persistence; 44 commits adding/updating tests in tests/clai2/browser; 5 commits adding/updating tests in tests/harness/daytona\_sandbox; 5 commits adding/updating tests in tests/harness/tool\_call\_judge; 55 commits adding/updating tests in tests/realtime/simulation; 60 commits adding/updating tests in tests/models/anthropic; 7 commits adding/updating tests in tests/harness/grain; 7 commits adding/updating tests in tests/harness/posthog; 7 commits adding/updating tests in tests/harness/repair\_tool\_arguments; 7 commits adding/updating tests in tests/harness/sprites\_sandbox; 8 commits adding/updating tests in tests/harness/ssh\_workspace; 82 commits adding/updating tests in tests/harness/code\_mode; 82 commits adding/updating tests in tests/harness/compaction; 89 commits adding/updating tests in tests/durable\_exec/temporal; 9 commits adding/updating tests in tests/ui\_roundtrip/known\_gaps; 97 commits adding/updating tests in tests/durable\_exec; Add lightweight inline\_snapshot stubs to speed up test suite startup; Add test asset files for dummy and product name content; Added Playwright browser smoke test and mutation testing script; Added benchmark suite for agent, tool search, and evaluation performance; Added comprehensive test coverage for the Shell capability; Added comprehensive test suite for pydantic-evals; Added comprehensive test suite for realtime sessions; Added comprehensive test suite for the FileSystem capability; Added comprehensive test suite for the GraphBuilder API; Added live MongoDB integration tests for step and media stores; Added live integration tests for LocalStack capability; Added live integration tests for PostgresStepStore and PostgresMediaStore; Added live integration tests for RedisSpendStore; Added regression tests for the gh-aw engine definition, version checks, and logfire isolation; Added test harness and coverage for the experimental ACP adapter; Added tests for BubblewrapSandbox harness capabilities; Added tests for ConversationSearch capability; Added tests for DynamicWorkflow capability and Temporal integration; Added tests for LocalStack container and toolset capabilities; Added tests for LogfireMCP capability and Temporal durability integration; Added tests for PydanticAIDocs capability and deprecation shim; Added tests for You.com search and research capabilities; Added tests for experimental deprecation shims; Added tests for graph utility functions and exception handling; Added tests for the CapabilityCreation harness and deprecated import shim; Added tests for the Coder harness capabilities; Added tests for the Day AI integration capability; Added tests for the E2B sandbox workspace capability; Added tests for the Macroscope code-review capability; Added tests for the ManagedPrompt capability; Added tests for the Memory capability and its storage backends; Added tests for the Notion hosted MCP capability; Added tests for the PromptInjectionDefender harness component; Added tests for the Pylon MCP capability integration; Added tests for the Researcher harness; Added tests for the Slack MCP capability; Added tests for the SpendLimits capability; Added tests for the StackOne capability and toolset; Added tests for the SubAgents capability; Added tests for the ToolOutputLimits harness capability; Added tests for the WarnOnCacheBusts harness capability; Added tests for the media storage harness; Added tests for xAI model integrations; Added unit tests for the OpenAI Codex provider; Major test suite overhaul and new mock infrastructure; New test infrastructure for the CLAI2 CLI; Provider test suite added to tests/providers.
Dependencies
Pydantic AI restructures into a multi-package workspace with new CLI and harness packages
The project has been reorganized into a \uv\ workspace containing several distinct packages: \pydantic-ai\ (the main framework, now a shim forwarding to \pydantic-ai-slim\), \pydantic-ai-slim\ (the core agent framework with optional provider extras like OpenAI, Anthropic, and Google), \pydantic-graph\ (a standalone graph/state machine library), \pydantic-evals\ (evaluation framework), \clai\ (a new command-line interface for chatting with LLMs), \pydantic-ai-harness\ (the official capability library and testing harness), and \pydantic-clai2\ (a capability-aware streaming terminal client). The main \pydantic-ai\ package now dynamically resolves its dependencies from \pydantic-ai-slim\ and exposes convenience extras (e.g., \cohere\, \mistral\, \temporal\) that map directly to the slim package's optional dependencies. Versioning is now handled dynamically via \uv-dynamic-versioning\ based on git tags, and the workspace enforces a minimum Python version of 3.10.
(dependencies) · high confidence
Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.
How this codebase got here
Score
- CAI 50 → 52 (+2.4)
- Rubric changed (rubric-2026.10.5 → rubric-2026.11.2) — scores are not directly comparable.
Lenses
- Code Health 65 → 63 (-2.0)
- Architecture 64 → 65 (+1.5)
- Maturity 52 → 82 (+30.9)
- Readiness 62 → 57 (-5.3)
- Security 79 → 76 (-2.8)
- Domain Modelling 100 → 100 (+0.0)
- Accessibility 38 → 39 (+0.7)
- Performance 94 → 94 (+0.0)
Resolved (313)
- AGUIEventStream._handle_tool_result (cognitive 18) (pydantic_ai_slim/pydantic_ai/ui/ag_ui/_event_stream.py)
- Advisor.init (cognitive 18) (src/pydantic_ai_harness/pydantic_ai_harness/advisor/_capability.py)
- Advisor.init (cyclomatic 16) (src/pydantic_ai_harness/pydantic_ai_harness/advisor/_capability.py)
- AgentRunResult._validate_serialized (cognitive 16) (pydantic_ai_slim/pydantic_ai/run.py)
- AnthropicModel._add_native_tools (cognitive 32) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._add_native_tools (cyclomatic 19) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._apply_message_cache_control (cognitive 18) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._get_betas_and_extra_headers (cognitive 19) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._get_betas_and_extra_headers (cyclomatic 19) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._get_container (cognitive 17) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._map_message (cognitive 339) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._map_message (cyclomatic 124) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._map_user_prompt (cognitive 23) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._messages_use_anthropic_uploaded_file (cognitive 29) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._prepare_tools_and_tool_choice (cognitive 36) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._prepare_tools_and_tool_choice (cyclomatic 20) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._process_response (cognitive 23) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicModel._process_response (cyclomatic 31) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicStreamedResponse._get_event_iterator (cognitive 117) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- AnthropicStreamedResponse._get_event_iterator (cyclomatic 68) (pydantic_ai_slim/pydantic_ai/models/anthropic.py)
- …and 293 more
New (404)
- 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)
- AnthropicModel._add_native_tools (cognitive 31) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._add_native_tools (cyclomatic 18) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._apply_message_cache_control (cognitive 19) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._get_betas_and_extra_headers (cognitive 22) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._get_betas_and_extra_headers (cyclomatic 22) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._get_container (cognitive 21) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._get_container (cyclomatic 16) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._map_model_response (cognitive 24) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._map_model_response (cyclomatic 18) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._map_user_prompt (cognitive 23) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._messages_use_anthropic_uploaded_file (cognitive 30) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._process_response (cognitive 32) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._process_response (cyclomatic 37) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicModel._tool_choice_declarations (cognitive 25) (pydantic_ai_slim/pydantic_ai/models/anthropic/init.py)
- AnthropicStreamedResponse._get_event_iterator (cognitive 132) (pydantic_ai_slim/pydantic_ai/models/anthropic/_streaming.py)
- AnthropicStreamedResponse._get_event_iterator (cyclomatic 73) (pydantic_ai_slim/pydantic_ai/models/anthropic/_streaming.py)
- BatchedModel._check (cognitive 19) (pydantic_ai_slim/pydantic_ai/models/_batch_model.py)
- BatchedModel._check (cyclomatic 17) (pydantic_ai_slim/pydantic_ai/models/_batch_model.py)
- …and 384 more
Changes since last survey
- 300 commits — 291 feature/other, 9 fixes
By area
- src/pydantic_ai_harness — 65 commits
- pydantic_ai_slim/pydantic_ai — 59 commits
- src/pydantic_clai2 — 56 commits
- tests/models — 18 commits
- .github/workflows — 14 commits
- tests/harness — 13 commits
- .github/scripts — 9 commits
- (root) — 7 commits
- docs/comparisons — 6 commits
- docs/harness — 6 commits
- .agents/skills — 5 commits
- tests/clai2 — 5 commits
- .github/api-compatibility-waivers — 3 commits
- docs/capabilities — 2 commits
- docs/durable_execution — 2 commits
- docs/ui — 2 commits
- pydantic_evals/pydantic_evals — 2 commits
- tests/AGENTS.md — 2 commits
- tests/conftest.py — 2 commits
- tests/durable_exec — 2 commits
Notable commits
- fix: Fix SubAgents under Temporal for a named agent with its own model and self with a model menu (#11862)
- fix: Fix web_fetch leaking script/style source into markdown content (#8329)
- fix: Fix a WarnOnCacheBusts test flake: record only the warnings its deprecation tests check, keeping the suite's filters for the rest (#12439)
- fix: Fix a resumed-run test flake: assert carried-over ToolReturnPart timestamps exactly instead of with IsNow (#12159)
- fix: Fix capability events emitted from tools under prefix_tools() (#11863)
- fix: Fix flaky SystemReminders caplog tests that counted other loggers' records (#11882)
- fix: Fix flaky coverage of the boxd test fake's post-exit output drain (#11842)
- fix: Fix instrumented embedding spans missing the embeddings attribute when include_content=True (#8295)
- fix: Let agents enqueue flake-fix PRs and re-enqueue removed PRs in the pushing-commits-to-the-repo skill (#12312)
- change: Accept Azure API-key authentication in Responses WebSocket tests (#12131)
- change: Accept list-typed content from OpenAI-compatible Chat Completions servers (#11885)
- change: Accumulate Mistral streamed tool-call deltas into one tool call (#11958)
- change: Add Azure Durable Task durability to pydantic_ai_harness (#11332)
- change: Add Edit and Retry to clai2 web UI prompts, rewinding like the TUI's rewind (#12220)
- change: Add WebSocket transport to OpenAI Responses (#10444)
- change: Add ConversationStore, which keeps a conversation as a log of messages stored once, and Conversation.revision to catch conflicting saves (#10658)
- change: Add ModelResponse.model_id recording the model the agent sent each request to (#11807)
- change: Add ModelResponsePartsManager.handle_tool_call_delta_events and forward every stream's tool call deltas through it (#12302)
- change: Add OpenSandbox for self-hosted agent workspaces (#11841)
- change: Add Pixeltable tools and memory storage to Harness (#9369)
- …and 280 more
Architecture
- Unchanged — 0 containers · 1 contexts · 0 edges
Written by cca.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.
Survey your own repository
pydantic/pydantic-ai was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.
About this page
- The score is its most recent published measurement, taken on 5 November 2026 at a pinned commit. It is not a live figure and does not change until the project is measured again.
- Measured at commit 8b455f222bc2dedc014aff970cbff3b502dc109c — the exact code this score is about.
- Scored under rubric-2026.11.2 — the same rubric and the same method as every other entry in this index.
- Measured by Canine Code Analyzer (cca.canine.dev) using cca measure-2026.11.2+preprod-4284b5b1cf64.