What’s New on GitHub Today – 12 Nov 2026

Here is today’s look at what developers are starring on GitHub, plus the latest product updates from the GitHub changelog. Star counts below are exactly as shown on GitHub Trending and its language pages (Python, JavaScript, TypeScript, PHP) on the evening of 12 November 2026 (IST). They change every hour, so expect different numbers when you click through.

The big theme: securing and measuring agent skills

Agent skills, the folders of instructions and scripts that coding agents load, have been all over Trending for weeks. Today the trend moved one step further: NVIDIA has two repositories on the Python list about checking skills, one for security and one for quality. If your team installs third-party skills into Claude Code, Codex or similar tools, these are worth a look.

Trending repositories

1. NVIDIA/SkillSpector

A security scanner for AI agent skills. Its description says it detects vulnerabilities, malicious patterns, prompt injection, data exfiltration and supply-chain risks in Claude Code, Codex and MCP skills before you install them.

• Language: Python
• Stars: 25,627 (302 today)
• Why devs care: skills can run scripts with your agent’s permissions. Scanning them before installation is the same habit as auditing npm packages, applied to a newer attack surface.

2. NVIDIA/SkillEvaluator

A multi-tier framework for evaluating agent skills, with quality gates, semantic overlap detection, synthetic evaluation dataset generation and live agent evaluation that measures how a skill changes agent behaviour.

• Language: Python
• Stars: 792 (83 today)
• Why devs care: once you write your own skills, you need a way to tell whether they actually help. Overlap detection is useful when several skills start competing for the same tasks.

3. CopilotKit/OpenBot

Open-source “AI coworkers” that each get their own computer: a browser, files and tools. The description says every action is decided before it happens and recorded after, and that it works with any AG-UI agent.

• Language: TypeScript
• Stars: 11,349 (291 today)
• Why devs care: the emphasis on pre-approved and recorded actions addresses the biggest worry about autonomous agents: knowing what they did and why.

4. QwenLM/qwen-code

An open-source AI coding agent that lives in your terminal, from the Qwen team.

• Language: TypeScript
• Stars: 29,641 (151 today)
• Why devs care: another open terminal agent to compare with commercial CLIs. Being open source means you can read exactly how it handles tools, files and context.

5. mksglu/context-mode

A free, local plugin that keeps tool output out of your coding agent’s context window, using a sandbox and SQLite FTS5 search. It lists support for Claude Code, Codex, Pi, OpenCode, Gemini CLI, Copilot CLI and more, and says it has zero telemetry.

• Language: JavaScript
• Stars: 33,961 (98 today)
• Why devs care: long test logs and command output fill an agent’s context fast. Storing output locally and letting the agent search it is a simple idea that can lower token costs.

6. OpenWhispr/openwhispr

A voice-to-text dictation app with local models (NVIDIA Parakeet and Whisper) and bring-your-own-key cloud models. It describes itself as privacy-first and cross-platform.

• Language: JavaScript
• Stars: 11,412 (135 today)
• Why devs care: dictating prompts, commit messages and docs is becoming a real workflow. Running speech recognition locally keeps what you say on your own machine.

7. ashhart/TensorFold

An LLM inference engine for Metal, CUDA and Vulkan.

• Language: Zig
• Stars: 1,868 (58 today)
• Why devs care: it is rare to see a Zig project on the overall Trending list. If you are curious how inference engines target several GPU APIs, a younger and smaller codebase can be easier to read than the big established ones.

8. Leantime/leantime

A goals-focused project management system for non-project managers, built with ADHD, autism and dyslexia in mind.

• Language: PHP
• Stars: 12,087 (26 today)
• Why devs care: it was among the top gainers on today’s PHP list, and is a good example of a self-hostable PHP app that puts accessibility at the centre of its design.

Quick table

GitHub changelog: latest updates

No new entries had appeared on the GitHub changelog for 12 November when we checked. The newest entry, from 11 November, is a useful one for individual maintainers. (The 10 November updates, including the compact dashboard and the C++ language server change, were covered in yesterday’s roundup.)

Issue types and fields for personal accounts (11 Nov)

Issue types and issue fields are now generally available on repositories owned by personal accounts, not just organisations. Here’s what changes:

• Issue types: the defaults are Bug, Task and Feature. You can customise them and create up to 25 from your personal account settings, on the Issue types page under “Planning”.
• Issue fields: the defaults are Priority, Effort, Start date and Target date. You can create up to 25, with text, number, date, single-select and multi-select types, and pin fields to specific issue types.
• Projects: types and fields can be used in projects for custom views and bulk updates.
• Automation: both are available through the REST and GraphQL APIs and webhooks.
• Labels and milestones from an issue: you can now create a label or milestone directly from the issue sidebar. If your search finds no match, GitHub offers to create one.

For solo open-source maintainers, this means the same structured triage organisations use. A simple setup is to add a “Priority” single-select and a “Bug / Feature / Task” type, then build a project view that sorts open bugs by priority.

Takeaway

Today’s list says the agent-skills ecosystem is maturing. People are no longer just sharing skills; they are scanning and evaluating them. If you install third-party skills, try SkillSpector before your next install. On the GitHub side, solo maintainers should turn on issue types for their personal repositories.