NEW 5 GitHub Repos Blowing Up (LIFE CHANGING)

Summary

The video reviews five GitHub repositories that gained more than 33,000 stars combined in one week. Each extends AI coding agents with workflows, memory, or a shared interface. The presenter ranks them by weekly star growth and recommends which ones are worth installing.

5. TradingAgents — 4,500+ new stars

TradingAgents is an open-source simulated trading research tool, not a system for placing real trades. Given a stock and date, it runs a multi-agent process: analysts review company records, Reddit, news, and charts; bullish and bearish agents debate; risk agents assess proposals; and a portfolio manager records a decision.

The presenter suggests examining historical debates rather than trusting the tool’s rating. He flags that the README’s “not financial advice” link was broken when checked, that performance claims rely on a short retrospective test, and that a single decision can involve dozens of model calls.

Verdict: Read the code, but don’t trade based on it.

4. Superpowers — nearly 6,000 new stars

Superpowers is a plugin that encourages coding agents to follow a more deliberate development process. Its skills prompt the agent to ask questions and draft a design before coding, use an isolated Git worktree, break work into small tasks, write tests first, and review each task. The presenter estimates that its always-active instructions use about 1,400 tokens.

The guide demonstrates installing it through Claude Code’s official plugin system and asking it to start a small React to-do-list project. Its main downside is overhead: the questioning can feel excessive for small fixes, and delegating tasks to sub-agents can consume substantial usage.

Verdict: The presenter’s top pick for changing an agent’s behavior.

3. ECC (formerly Everything Claude Code) — 6,000+ new stars

ECC is a large collection of Claude Code configuration: 71 agents, 314 skills, 98 commands, and 27 hooks. Setup offers minimum, standard, and strict hook profiles. Hooks can load context, block risky commands, and run checks such as formatting and type checks. Specialist agents include planners and code or security reviewers. Optional background features can record “instincts” and evolve them into skills.

The presenter recommends treating ECC as a reference library rather than installing everything. He estimates that exposing descriptions for all the skills could add roughly 30,000 tokens. The README also warns against unofficial re-uploads because they may contain malware.

Verdict: Choose a few useful skills or minimal hooks, and avoid stacking it with Superpowers.

2. claude-mem — 7,000+ new stars

claude-mem is a plugin intended to carry context between Claude Code sessions. Hooks monitor tool calls, and a low-cost model such as Haiku writes brief notes that are stored locally in SQLite and a vector database. A later session starts with a summary of recent notes and can search for more detail.

The presenter highlights a serious privacy concern in the project’s release notes: a browser-based installation path uploaded the full text of recent prompts, along with session and project metadata, without an on-screen consent prompt. The project later added a consent screen and a way to keep data local, but says previously uploaded prompts are not removed. The presenter also notes that analytics are enabled unless disabled, and that the note-taking model uses the user’s plan. The project’s notes associate the reported upload with browser setup, not the plugin installation path.

Verdict: Promising, but wait for the updated approach to prove itself and review telemetry settings before trying it.

1. T3 Code — nearly 10,000 new stars

T3 Code is a free, open-source desktop, web, and phone interface for managing coding agents, including Claude Code, Codex, Cursor, Grok Build, and Antigravity. Each task can have its own Git worktree, diff, preview, and persistent queue. A delegated-task feature can have one agent plan work and hand implementation to another; T3 Connect lets users monitor work remotely.

The tutorial suggests signing into the agents, adding a project, and then trying a workflow such as having Claude plan and Codex build. The presenter cautions that the project describes itself as very early and has many open issues. New threads default to full access, Claude through the Agent SDK may use up five-hour limits faster than Claude Code directly, and usage analytics are enabled by default.

Verdict: Useful if you pay for multiple agents, but change the access setting and review analytics first.

Overall Recommendations

• Install first: Superpowers, if you want an agent to plan and test more systematically.
• Consider: T3 Code if you use multiple coding agents; use a spare machine and adjust its defaults.
• Use selectively: ECC—pick a few skills or hooks rather than installing the whole bundle.
• Wait: claude-mem, given its recent data-handling changes.
• Treat as research only: TradingAgents; don’t rely on it to make trades.

Main Speakers and Sources

• Main speaker: The video’s narrator, from the Beau Johnson channel.
• Sources discussed: The repositories’ READMEs, documentation, issues, FAQs, release notes, and screenshots. The narrator also refers to statements by project creators and users.