GitHub engineering blog

Loading
Building the Agent Host Protocol for sessions that continue across devices
GitHub describes the Agent Host Protocol (AHP), an open standard that keeps Copilot agent sessions continuous across VS Code, the GitHub Copilot app, mobile apps, and cloud sandboxes. The post explains why per-client integrations failed, how a URI-channel, reducer-based state design handles reconnection and concurrent mutations, and what was learned from building two independent hosts in TypeScript and Rust. It also covers how the protocol stays agnostic to agent behavior and where to connect...
Maintainers don’t need more AI bug reports. They need fixes.
GitHub and Microsoft ran a pilot using Codename MDASH, an AI scanning harness, to find vulnerabilities in six open source projects. An AI triage agent removed duplicates and researchers verified each finding by hand before reports went out privately with proofs of concept and suggested fixes. The post distills lessons for anyone sending AI-generated security reports: verify findings, report privately, and include a fix.
GitHub availability report: October 2026
GitHub's October 2026 availability report covers eight incidents, including Actions job-assignment failures caused by a cloud networking disruption, a migration-driven database overload, a write outage from a misrouted database leader, and a Copilot credential failure. It also details ongoing Azure capacity expansion and reliability work, with each incident broken down by root cause, response, and planned mitigations.
Build once, run on any agent: A practical guide to agent skills
This GitHub Blog guide explains how to write portable agent skills as versioned SKILL.md directories that capture repeatable engineering work, while tools, credentials, and permissions stay with the host. It walks through the Proof Agent verifier pattern, the Data Forge multi-skill pipeline with a shared manifest and common platform adapter, human checkpoints, CI-based checks, and a packaging script that generates client-specific plugin files from one source.
Git security vulnerabilities announced
The Git project released versions 2.56.1 and several backported maintenance releases to fix six security vulnerabilities, including code execution during recursive clones via an index path truncation bug, path traversal through a server's advertised HEAD symref, stack corruption from merging tags with many GPG signatures, out-of-bounds reads in binary patch handling, heap corruption in large merges, and filesystem access via fast-import features. The post details each flaw's mechanism and fix...
Beyond code generation: A developer’s guide to agentic engineering
GitHub outlines an agentic engineering workflow across planning, building, verifying, and observing software work, using Copilot features such as Plan Mode, cloud-sandboxed sessions, Copilot code review, AI Scan, and Agentic Autofix. The post highlights pull request review bottlenecks and post-release investigation, but mostly presents GitHub's own product capabilities rather than implementation detail.
Accessibility statements now have a home on GitHub
GitHub now recognizes ACCESSIBILITY.md files in a repository's root, .github/, or docs/ directory, shows them on repository overview pages, and lets repositories inherit a statement from their organization. The post outlines recommended sections for an accessibility statement, including supported environments, known limitations, contributor expectations, and reporting processes, and notes that a WCAG target is not the same as verified conformance.
Inside GitHub’s new Git storage architecture
GitHub describes its new Git storage architecture, called Rubicon, which separates reference state into a dedicated replicated log (Ref-CAS) and keeps Git objects in Azure Blob Storage so that thousands of agents can push to one repository. The post explains why object-storage-based reference designs hit write ceilings, how per-reference compare-and-swap with LSN-based version checks keeps reads cheap, and how background compaction with leased shards keeps pack counts flat under heavy write...
Arnis turns any place on Earth into a Minecraft world
Louis Erbkamm describes Arnis, an open source project that turns OpenStreetMap and other geospatial data into explorable Minecraft worlds. The post outlines the generation pipeline, the Python-to-Rust port, and how he uses GitHub Copilot sub-agents to review large pull requests and catch regressions. It is a vendor-written maintainer profile with limited technical depth.
Game Off 2026: The theme is SHIFT
GitHub announces Game Off 2026, its annual month-long game jam, with the theme SHIFT and a submission deadline of December 1, 2026. The post covers participation steps, voting categories, and a list of suggested game engines, but offers little beyond event logistics and no transferable technical insight.
GitHub Universe 2026: Evolving every layer of the stack
GitHub outlines a vision for an agentic software delivery system and announces Copilot app, Copilot Fusion multi-model orchestration, Copilot projects, session continuity, automations, and a Max tier for code review. The post also cites a benchmark comparison, capacity additions, a planned multi-region Azure migration, and three principles behind a new Git architecture.
Octoverse: Commits, pushes, reach new highs in 2026 as developers build with agents
GitHub's 2026 Octoverse report finds 17.9 billion code pushes over the past year, with agent-started pull requests rising from 17% in March to over 40% by September. It also examines Copilot model-selection patterns, language growth led by TypeScript and Python, and industry breakdowns of public repositories, using a documented methodology built on GitHub's own telemetry.
How bug bounty sharpens one researcher’s security skills
GitHub's Bug Bounty team profiles researcher @r31n, who focuses on Enterprise Server. The interview covers source-and-sink tracing, exploit chaining such as combining an open redirect with an allowlisted SSRF, and why AI output and model guardrails should be treated as untrusted input rather than as a security boundary.
The bridge to agent-scale Git infrastructure
GitHub describes how it moved most of its repository replicas to Azure and how changing its Spokes quorum from repository-wide to per-reference serialization removes the write bottleneck caused by cross-region latency. The post also explains how object storage acts as a virtual replica to decouple reads from the write quorum, enabling independent read scaling. Internal tests reported 3-4x higher write throughput for writes to different references.
How 3 Java Champions are using GitHub Copilot for increased productivity
Three Java Champions describe how they use GitHub Copilot in production and open source work. They cover an agent-driven migration verified with screenshots and PDF diffs, a virtual thread pinning bug in a native poll() call that stalled the fifth input watcher, and reusable agent skills paired with deterministic quality checks. The post emphasizes keeping humans accountable for what ships.
How to continue a GitHub Copilot app session in VS Code
This GitHub Blog tutorial explains how to open an agent session from the GitHub Copilot app in VS Code, keeping its conversation, diff, and worktree intact. It walks through reviewing the agent's changes, editing and testing locally, and pushing commits that update the associated pull request.
Improving site performance by shipping more CSS
GitHub details its multi-year migration from CSS-in-JS to CSS Modules across Primer and github.com. The post explains the incremental feature-flagged rollout, automated sx-prop conversions, theming changes, and measured gains including 55% faster server-side rendering and 25% faster component initialization.
Staying Evergreen: How we automated secure access and repo hygiene in a shared org
GitHub describes Evergreen, a GitHub App built from native features (scheduled Actions, repository_dispatch, webhooks, issues, custom properties, and git-tracked JSON state) that runs quarterly outside-collaborator access reviews and stale-repository archival in a large shared org. The post covers fail-secure and fail-closed design, issue-based IssueOps with privileged-closer checks, owner-resolution fallback chains, dry-run rollout, and pacing against secondary rate limits.
The next developer interface is ambient intelligence
The GitHub Blog post argues that AI agents, voice, and physical controls are moving developer tooling from issuing commands to expressing intent. It introduces a three-layer model (modality, harness, control plane) and describes a Logitech MX Keypad integration with GitHub Copilot. The post is mostly conceptual and promotional, with little concrete implementation detail.
From 2,500 requests to 24: Rebuilding star history without the stargazers
GitHub's engineering team describes replacing the stargazers-based star history workflow with an aggregate endpoint that returns up to 30 weeks of daily star counts without identities. The post explains why the original endpoint was costly, including a pagination bug that counted all stars on every request, and how filtering by indexed created_at range before grouping cut requests for react/react from about 2,511 to 24. It also covers remaining limits such as dense weeks and calendar boundaries.