Pull Request Explorer

Summary

Fix linting and review comments for hosting

Improves code quality and addresses review feedback, ensuring reliable hosting setup.

Health Assessment

• • Fast review and merge with minimal changes, low risk to production.

Summary

Release version 0.105.2

Automated release of SDK version 0.105.2, updating changelog and version file.

Health Assessment

• • Release PR merged immediately with no review, minimal code changes.

Summary

Release version 0.105.1

Automated release of SDK version 0.105.1, updating metadata and publishing to PyPI.

Health Assessment

• • Release PR completed in minutes with minimal changes and no review friction.

Summary

Add hosting cookbook for Docker, Modal, Kubernetes

Provides developers with a ready‑to‑run reference implementation for deploying Claude agents across Docker, Modal, and Kubernetes, accelerating deployment and reducing setup time. This empowers teams to quickly prototype and scale agent deployments.

Health Assessment

• • Fast review and single commit indicate a straightforward change; large scope due to notebook and infrastructure code but low risk as it is a documentation and reference implementation.

Summary

Release SDK 0.105.0 with new model support and file size caps

Adds support for the Claude‑opus‑4‑8 model, mid‑conversation system blocks, detailed usage tokens, and custom file size caps, enhancing API flexibility and developer experience.

Health Assessment

• • The PR had a long cycle time with multiple commits and a revert, indicating significant rework and review friction.

Summary

Migrate Claude workflows to Workload Identity Federation

Switches 6 CI workflows from static API key to secure Workload Identity Federation, improving credential security and compliance.

Health Assessment

• • Fast review and merge, minimal changes, low risk.

Summary

Release version 0.104.1 with streaming bug fix

Fixes a bug in the streaming beta compaction accumulator that preserved encrypted content, improving data integrity for downstream consumers.

Health Assessment

• • PR merged within 18 hours with a single change, indicating a straightforward bug fix and release with minimal review friction.

Summary

Add Tier 3 Kubernetes hosting with pod‑per‑session gateway

Provides a Kubernetes‑based hosting tier for the Agent SDK, enabling pod‑per‑session gateways, egress lockdown, Redis routing, and a quickstart environment.

Health Assessment

• • PR was merged within 1.6 hours with only one commit after review, but required three rounds of changes, indicating a complex but well‑managed implementation of a new Kubernetes hosting capability.

Summary

Add support for thinking-token-count beta

Adds a new beta feature that estimates token counts in thinking block deltas during streaming, improving accuracy for token usage metrics.

Health Assessment

• • Rapid release with minimal changes and quick review indicates low complexity and high confidence in the new feature.

Summary

Pin GitHub Actions to commit SHAs

Pin third-party GitHub Actions references to immutable commit SHAs.

Health Assessment

• • PR was straightforward with quick review, but long cycle time indicates backlog.

Summary

Pin pre-commit hooks to immutable commit SHAs

Ensures reproducible and secure CI pipelines by locking pre-commit hooks to specific commit SHAs, preventing accidental upgrades that could introduce bugs or vulnerabilities.

Health Assessment

• • Long review cycle indicates low priority or low complexity; minimal changes and quick merge after review.

Summary

Bug fix: SessionToolRunner skip tool calls

Fixes a bug where SessionToolRunner incorrectly attempted to call tools it does not own, improving reliability of self-hosted managed-agent sessions.

Health Assessment

• • Rapid merge with minimal review friction indicates a straightforward bug fix with low impact on the codebase.

Summary

Rename private sandbox to self-hosted sandbox

Aligns naming across documentation and deployment artifacts to match the self-hosted sandbox feature, improving consistency and reducing confusion.

Health Assessment

• • Quick rename with immediate approval, minimal rework, high confidence in consistency.

Summary

Add self-hosted sandbox worker templates

Provides reference implementations for running Claude Managed Agents on self-hosted sandboxes across multiple compute providers, enabling secure, environment-key-only execution.

Health Assessment

• • Rapid review and merge within 3.5 hours indicates high confidence and minimal friction.

Summary

Add self-hosted sandbox support to CMA

Enables users to run models in self-hosted sandboxes, expanding deployment flexibility.

Health Assessment

• • Fast cycle time and single review round indicate smooth integration.

Summary

Slack stateless webhook bridge template

Provides a stateless TypeScript/Bun bridge connecting Slack Events API to Claude Managed Agents via CMA outbound webhooks, enabling restart‑safe, horizontally scalable integration without a session‑map database.

Health Assessment

• • PR had a single commit and quick approval, indicating straightforward implementation with minimal review friction.

Summary

Add Linear stateless webhook bridge template

Provides a lightweight, stateless bridge between Linear and Claude Managed Agents, enabling scalable webhook integration without maintaining session state. This simplifies deployment and reduces operational overhead for teams using Linear's Agent Platform.

Health Assessment

• • The PR required over five days to receive its first review and merge, indicating a slow review cycle. The large addition of 540 lines across 12 files suggests significant changes that could impact stability and warrant careful testing.

Summary

Release 0.102.0

Automated release with new features and spec updates for the anthropic-sdk-python

Health Assessment

• • Automated release PR with multiple commits and changes
• • Long cycle time due to automated process

Summary

Add CMA Sessions API MCP server

This PR introduces a lightweight MCP server that allows Claude Desktop and claude.ai to interact with hosted CMA agents via tool calls. It provides both stdio and HTTP interfaces, enabling seamless integration until official support is available.

Health Assessment

• • The PR introduced a substantial new server component with a large code addition, but the review process was relatively straightforward with a single commit and no major rework.

Summary

Release version 0.101.0 with updated examples

This release updates example scripts to use uv, bumps tool references, adds an AWS client for Claude Platform, and fixes a client error message. These changes improve developer experience and expand platform support.

Health Assessment

• • The PR had a long review cycle but involved only minor changes, indicating low technical risk but potential process delays.

Summary

Retag managed_agents notebooks with Claude category

Updates registry to use new 'Claude Managed Agents' category for managed_agents notebooks, ensuring correct pill display on platform.

Health Assessment

• • Quick turnaround with minimal changes; likely low risk.

Summary

Bump example model to latest Claude

Align the tools_runner example to use the current Claude model, ensuring users copy‑paste the correct model identifier.

Health Assessment

• • The PR had a long review cycle but involved only a single line change, posing minimal risk to the codebase.

Summary

Update example shebangs to uv

Ensures example scripts run without poetry, aligning with repo's uv migration.

Health Assessment

• • Small change, long review time due to scheduling, no rework needed.

Summary

Fix docs rendering and rename multiagent tutorial

Notebook outputs now render properly on the docs site, improving user experience. Tutorial titles are standardized for consistency across the cookbook.

Health Assessment

• • Quick turnaround with minimal review indicates a straightforward bug fix.

Summary

Add multiagent and outcomes cookbooks

Adds two Managed Agents cookbooks demonstrating multiagent coordination and outcome-based grading, enhancing user ability to build complex agent workflows.

Health Assessment

• • PR merged in under an hour with a single review comment, indicating a straightforward addition of example notebooks with minimal friction.

Summary

Release version 0.100.0 of SDK

Adds support for Managed Agents multiagents, outcomes, webhooks, and vault validation, improving API capabilities.

Health Assessment

• • Release completed within 20 hours with a single review cycle, indicating an efficient process but the large volume of changes increases risk.

Summary

Release SDK version 0.99.0

Bumps the SDK to 0.99.0, adding support for targeting a workspace when exchanging OIDC federation tokens to improve client flexibility.

Health Assessment

• • Rapid release with minimal review and changes, indicating a straightforward update with a single new feature.

Summary

Add vulnerability detection agent cookbook

Provides a step‑by‑step guide for building a vulnerability detection agent with Claude SDK, enabling threat modeling, find, triage, and reporting on a C target.

Health Assessment

• • Fast cycle time and minimal review iterations indicate a smooth process; AI-assisted code generation helped deliver the cookbook quickly.