Copilot GA, ARC-AGI, and Infra Updates #94

July 31, 2026 — 2 min read

Today's Letter

GitHub made agent skills and MCP server support generally available in Copilot code review on July 29, 2026
The feature is available to Copilot Pro, Pro+, Business, and Enterprise users
Agent skills let code review invoke repository or organization-specific instructions from SKILL.md files under .github/skills
MCP server connections can pull review context from external systems such as issue trackers, documentation platforms, and service catalogs
GitHub says all MCP tool calls made during code review are restricted to read-only access
Existing MCP configurations for Copilot cloud agent carry over to Copilot code review without additional setup
GitHub and Playwright MCP integrations are enabled by default
Code review comments now show attribution when they were generated with agent skills or MCP-based context

Source: github.blog

Cloudflare published a post describing the migration of cdnjs to its Developer Platform
The write-up frames the work as large-scale internal dogfooding rather than a separate product launch
cdnjs appears as the production workload used to validate Cloudflare's developer stack under real traffic
The post is filed under Developers and tagged with CDNJS, Cloudflare Workers, Durable Objects, R2, Queues, and Containers
The update indicates continued investment in running long-lived public infrastructure on Cloudflare's own platform services

Source: blog.cloudflare.com
More: anthropic.com

OpenAI said two runtime settings raised GPT-5.6 Sol's ARC-AGI-3 public-set score from 13.3% to 38.3%.
The same setup reduced output tokens by 6x versus the official benchmark harness.
The two changes were retained reasoning and compaction, both already used in ChatGPT and Codex.
In the official harness, private reasoning was discarded after each action and older history was lost through rolling truncation.
OpenAI rebuilt the harness on the Responses API, where passing the previous response ID retains reasoning across tool calls and turns.
OpenAI said retained reasoning reduced per-turn deliberation and improved strategy learning over longer runs.
GPT-5.6 Sol previously scored 7.8% on ARC-AGI-3, while GPT-5.5 scored 0.4%.
OpenAI estimated the average human tester score at 48% from official gameplay logs.

Source: openai.com

AWS published an official post on 2026-07-29 about Private Key JWT authentication with Amazon Bedrock AgentCore Identity.
The flow centers on client authentication with signed JWTs rather than a shared OAuth 2.0 client secret.
The referenced AWS services are Amazon Bedrock AgentCore Identity, Amazon Bedrock, and Amazon Bedrock Agents.
The material is framed as an implementation pattern for agent identity and authentication, not as a broader platform release note.
Cross-checking sources did not independently confirm the same claims, but the primary source is the official AWS domain and was treated as verified.

Source: aws.amazon.com

Jocoletter curates AI, software, and product trends for developers and builders.

#AWS #Cloudflare #GitHub #OpenAI