GitHub Trending Weekly 2026-08-14: Agents Gain Computers, Memory, and Workflows
Data period: 2026-08-06 to 2026-08-14 (rolling 7 days)
Sources: GitHub Trending weekly and monthly, GitHub Search API, GitHub REST API, HN Algolia
TL;DR: prime-agent led this week's fastest-growing list, followed by Cloudflare Computer and TencentDB-Agent-Memory. The new-repository list brought DeepSeek Harness, phone control, offline file conversion, and local Mac inference. The shared signal is clear: open-source attention is moving from what an agent can answer to whether it can remember, operate, hand off, and keep working safely.
Fastest Growing: Top 15
🔁 means the project also appeared in the monthly trend list. All numbers come directly from the fetched JSON and are not a quality guarantee.
| # | Project | Stars/week | Total stars | Language | Created |
|---|---|---|---|---|---|
| 1 | prime-agent | +12,612 | 15,331 | TypeScript | 2026-05-08 |
| 2 | computer | +6,020 | 7,940 | TypeScript | 2026-06-05 |
| 3 🔁 | TencentDB-Agent-Memory | +5,720 | 21,110 | TypeScript | 2026-04-07 |
| 4 🔁 | reverse-skill | +5,573 | 24,820 | PowerShell | 2026-05-13 |
| 5 | agent-skills | +4,817 | 86,810 | JavaScript | 2026-02-15 |
| 6 | pdf-inspector | +4,043 | 15,257 | Rust | 2026-02-06 |
| 7 🔁 | book-to-skill | +3,983 | 21,125 | Python | 2026-05-01 |
| 8 | semantica | +3,585 | 6,310 | Python | 2025-06-25 |
| 9 | ComfyUI | +3,321 | 127,289 | Python | 2023-01-17 |
| 10 | DeepSeek-Reasonix | +2,953 | 34,426 | Go | 2026-04-21 |
| 11 | loopx | +2,509 | 4,578 | Python | 2026-05-31 |
| 12 | ChinaTextbook | +2,384 | 79,248 | Roff | 2020-01-05 |
| 13 | google/skills | +2,288 | 18,029 | Python | 2026-03-31 |
| 14 | code-graph-rag | +1,511 | 4,177 | Python | 2025-06-16 |
| 15 | manim | +1,420 | 90,743 | Python | 2015-03-22 |
Top New Repos: Top 15
| # | Project | Total stars | Language | Created |
|---|---|---|---|---|
| 1 | deepseek-harness | 33,063 | TypeScript | 2026-08-13 |
| 2 | watermarks-remover | 4,537 | Python | 2026-08-11 |
| 3 | flyingmouse-format | 2,359 | JavaScript | 2026-08-06 |
| 4 | h3.c | 1,753 | C | 2026-08-09 |
| 5 | phone-harness | 1,706 | Python | 2026-08-07 |
| 6 | oil-motion | 1,674 | Python | 2026-08-07 |
| 7 | WeChat-AI | 1,671 | TypeScript | 2026-08-10 |
| 8 | shuohao-skills | 1,281 | JavaScript | 2026-08-06 |
| 9 | Dalaran | 930 | Rust | 2026-08-07 |
| 10 | kage | 884 | HTML | 2026-08-08 |
| 11 | youtube-music | 855 | TypeScript | 2026-08-11 |
| 12 | pi-from-scratch | 846 | TypeScript | 2026-08-09 |
| 13 | claudish-to-english | 846 | Shell | 2026-08-10 |
| 14 | asm-hall-of-shame | 836 | C | 2026-08-06 |
| 15 | OpenMausBot | 782 | TypeScript | 2026-08-11 |
What stood out
prime-agent describes itself as a self-improving RLM agent for coding workflows and long-running autonomous tasks. It gained 12,612 stars this week and reached 15,331 total. The useful question is not whether it can produce one impressive result, but whether it can iterate safely over a long task.
Cloudflare Computer points agents toward operating a computer. TencentDB-Agent-Memory turns conversations, documents, and code into shared memory assets. Together with loopx, they make continuity the headline: can the agent resume yesterday's work, show evidence, and stop safely when something goes wrong?
reverse-skill, agent-skills, book-to-skill, and google/skills show reusable skills becoming workflow components. pdf-inspector is a practical document layer, while ComfyUI remains a major visual workflow entry point. semantica and code-graph-rag focus on context, graphs, provenance, and codebase understanding rather than another chat surface.
The new list pushes closer to devices and daily work. phone-harness advertises phone control, flyingmouse-format targets offline Windows conversion and OCR, h3.c targets local Mac inference, and pi-from-scratch is a compact way to study an agent loop. watermarks-remover, youtube-music, Dalaran, and kage cover provenance, media, robotics visualization, and interactive storytelling. Treat new stars as an attention signal, not proof of maturity, especially when a tool touches accounts, files, or devices.
Monthly trend comparison
The projects appearing in both weekly and monthly lists are TencentDB-Agent-Memory, reverse-skill, and book-to-skill. All three relate to reusable agent capabilities, suggesting sustained attention rather than a single-day spike. A weekly-only project is not necessarily weaker; it simply reflects shorter-term acceleration in this snapshot.
This week's insights
The agent competition is moving from answers to continuity. Computer control, memory, durable goals, and handoffs matter when automation has to run beyond one prompt.
Skills are becoming a reusable workflow format. That can reduce repeated setup, but permissions, provenance, and cross-platform compatibility become essential checks.
The practical layer still matters. PDF inspection, local inference, image workflows, and device data often remove one real bottleneck faster than a more ambitious agent framework.
Closing thought
This week's list feels like a shift from “build another chatbot” toward giving agents hands, memory, state, and a way to pass work forward. If you want an approachable starting point, look at pdf-inspector, ComfyUI, or pi-from-scratch. For longer automation, study memory, skills, and computer use, but do not let a single week's star count make the decision for you.
Was this article helpful?
