GitHub listed fifteen projects this Wednesday morning, and ten of them were featured here in the past week, so most stay out. Today's five picks: MiniMind, a from-scratch course in training a small language model; Cursor's official plugin repository; Tencent's LoopForge, back after a security-minded 0.4.0 release; Superpowers, back after a release that reworks how it asks questions; and Svelte, whose 5.58.0 release landed this morning. Curation order reflects practical interest, not GitHub rank. Trending is not a release date or an endorsement.
Picks are listed in curation order, not GitHub’s rank; each card shows the rank GitHub displayed when the page was checked. Stars measure attention, not security, quality or fitness for use, and no project here was installed or run. “Stars today” is GitHub’s own displayed figure for its daily window, not an audited rolling 24-hour count.
Pick 1 · AI / machine learning educationGitHub rank #14 of 15
What it does. A from-scratch course in building a small language model: native PyTorch code and open data for every stage, from tokenizer and pretraining through fine-tuning, DPO, GRPO-style reinforcement learning and distillation, ending in a 64M-parameter chat model.
Caveat. The headline two hours is, by the README's own note, one epoch of the SFT stage on a single NVIDIA 3090, not the whole pipeline, and the RMB 3 figure is the matching GPU rental cost; both are maintainer claims. A 64M model is a teaching tool, not a capable assistant, and the main README is in Chinese with an English translation.
Attention, not endorsement
GitHub showed
“192 stars today” as displayed for GitHub’s daily window; not an audited rolling 24-hour count
What it does. Cursor's official plugin repository: the plugin specification plus first-party and partner plugins that bundle agent skills, rules and MCP server definitions, from code-review and orchestration helpers to Gmail, GitHub and Salesforce connectors.
Caveat. GitHub detects no license for the repository: the README ends with an MIT line and the plugin folders we sampled carry an MIT LICENSE file, but there is no root license, so check each plugin. Connector plugins give an agent access to mail, documents and business data under your account, so review their scopes first. There are no releases or version tags.
Attention, not endorsement
GitHub showed
“201 stars today” as displayed for GitHub’s daily window; not an audited rolling 24-hour count
What it does. A self-hosted workspace from Tencent for shipping software with coding agents: a chat bot opens Docker sandboxes with your repositories, delegates to project roles and opens pull requests, and proven workflows become SOPs of agent, script and human-approval steps.
Caveat. It is still pre-1.0 software: 0.3.9 will never get a final release, and existing installations must apply new database migrations themselves. It runs as a Docker stack with MySQL, each executor needs its own model credentials, and the security fixes are the maintainers' description, not something we audited.
Attention, not endorsement
GitHub showed
“591 stars today” as displayed for GitHub’s daily window; not an audited rolling 24-hour count
What it does. An agentic skills framework and development method for coding agents: a library of skills that walks an agent through brainstorming, planning, test-driven implementation and review, installed separately for Claude Code, Codex, Cursor, Copilot CLI and other harnesses.
Caveat. The blind-graded improvement figures come from the maintainers' own test scenarios, not an independent benchmark. It works by adding session-start hooks and persistent instructions to your agent, so review what each harness install changes; the optional visual companion has telemetry you can switch off.
Attention, not endorsement
GitHub showed
“1,917 stars today” as displayed for GitHub’s daily window; not an audited rolling 24-hour count
What it does. A compiler-based UI framework: Svelte turns declarative components into JavaScript that updates the DOM directly, and is developed by volunteers under the MIT license.
Caveat. This is an incremental release, not a new major version: the two features target devtools and dev-mode metadata, and the performance notes carry no figures. GitHub does not say why a project is trending, and nothing shown ties today's attention to this release.
Attention, not endorsement
GitHub showed
“477 stars today” as displayed for GitHub’s daily window; not an audited rolling 24-hour count
Every repository GitHub listed, in GitHub’s order, with the figures it displayed when the page was checked (): 5 picked today, 10 covered in an earlier digest, 0 not covered. For projects covered before, growth compares the total stars at our first pick with today’s displayed total.
A Claude Code skill bundle for bug hunting and external red-team work - 82 skills, 15 slash commands, 681 disclosed-report patterns curated across 24 core vulnerability classes, plus enterprise identity + infrastructure attack matrices.
Covered before — first picked 16 Nov 2026 at 6,748 stars; now 9,864: +3,116 stars (+46.2%) in 2 days, about 1,558 a day.
Self-hosted workflows for coding-agent delivery. Orchestrate CodeBuddy, Codex and Claude Code with scripts and human approval gates. Includes Local workflows for coding assistants.
Picked today — Pick 3 above; first covered 15 Nov 2026 at 1,016 stars; now 3,521: +2,505 stars (+246.6%) in 3 days, about 835 a day.
TypeScript · 1,774 stars today · 29,423 stars in total
Build your own AI office with Univer Office SDK. Bring editing, calculation, collaboration, and file processing into your product for people and AI agents. Built on an extensible open-source core.
Covered before — first picked 15 Nov 2026 at 25,093 stars; now 29,423: +4,330 stars (+17.3%) in 3 days, about 1,443 a day.
TypeScript · 350 stars today · 10,203 stars in total
A library of Agent Skills designed to work with the Stitch MCP server. Each skill follows the Agent Skills open standard, for compatibility with coding agents such as Antigravity, Gemini CLI, Claude Code, Cursor.
Covered before — first picked 16 Nov 2026 at 9,464 stars; now 10,203: +739 stars (+7.8%) in 2 days, about 370 a day.
Shell · 1,917 stars today · 319,545 stars in total
An agentic skills framework & software development methodology that works.
Picked today — Pick 4 above; first covered 17 Oct 2026 at 300,784 stars; now 319,545: +18,761 stars (+6.2%) in 32 days, about 586 a day. Also picked 28 Oct 2026.
TypeScript · 691 stars today · 13,178 stars in total
The workspace to stay in the loop and on top of agents - shape, decide, plan, annotate, review code, share with team, and send all feedback to agents with one click.
Covered before — first picked 14 Nov 2026 at 10,567 stars; now 13,178: +2,611 stars (+24.7%) in 4 days, about 653 a day.
JavaScript · 883 stars today · 305,814 stars in total
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Covered before — first picked 16 Oct 2026 at 280,607 stars; now 305,814: +25,207 stars (+9.0%) in 33 days, about 764 a day.
PaperOrchestra is a multi-agent framework for automated AI research paper writing.
Covered before — first picked 17 Nov 2026 at 1,824 stars; now 3,385: +1,561 stars (+85.6%) in 1 day.
Following Our Picks
Every project picked in an earlier digest, biggest growth first. “Then” is the total at our first pick’s metadata check; “now” is GitHub’s API star count checked . Growth measures attention, not quality or safety.
GitHub's daily listing held 15 rows at capture. The capture parsed exactly fifteen listing rows and is GitHub's own list, not a truncation by us.
All 15 listing entries were inspected. Exact public GitHub API metadata of the five not-recently-featured entries was read between 07:01:17 and 07:01:23 UTC and of Tencent/LoopForge at 07:02:32 UTC; README files between 07:01:46 and 07:02:10 UTC; recent releases of all fifteen entries just after 07:01:23 UTC. Every checked repository was active and unarchived.
Ten entries featured in the past seven days were checked for releases since their cards. Nine were left out: alibaba/open-code-review and backnotprop/plannotator (14 November), dream-num/univer and ag-ui-protocol/ag-ui (15 November), elementalsouls/Claude-BugHunter and google-labs-code/stitch-skills (16 November), and getarcaneapp/arcane, mvanhorn/last30days-skill and google-research/paper-orchestra (17 November). Their releases since those picks are patch builds or dated routine releases, or none at all. Tencent/LoopForge returns because of its 0.4.0 release, as its card explains.
affaan-m/ECC was inspected and not picked: it was featured on 16 October and its latest release, v2.2.3 of 1 October, predates that card, so no material change was identified.
obra/superpowers was last featured on 28 October and returns because of its v7.0.2 release of 3 November, as its card explains.
Tencent/LoopForge shows NOASSERTION in GitHub's license detection; its LICENSE file was read directly and states MIT for LoopForge, with bundled third-party components under their own licenses.
cursor/plugins has no license detected by GitHub and no root LICENSE file. Its README ends with an MIT line, and the three plugin folders sampled (teaching, thermos, orchestrate) each carry an MIT LICENSE file; other plugins were not checked one by one.
Disclosure: The AI Gazette's twice-daily research step uses an earlier version of the last30days engine (mvanhorn/last30days-skill), which was on today's list at rank 2 and was featured yesterday.
The preceding seven calendar dates, 11-17 November, were checked for local history; all seven editions exist.
The official listing was captured at 08:00:57 Europe/Paris. Literal stars-today labels are GitHub's displayed period figures; API total stars and README evidence were checked separately at the times shown per card. They are not an independently audited rolling 24-hour growth measurement.
Feature descriptions, adoption and performance figures come from each project's own README and release notes and are the maintainers' claims. No candidate code, installer, CLI, plugin, skill or model was installed, downloaded or run, and none of this is a security audit or endorsement.