How agency-agents Gained 11K Stars in a Week by Embedding 140 Expert Roles into Your Editor

AI Agents Are No Longer Toys

In the past week, GitHub’s trending data highlighted a clear signal: AI agents have moved from experimental "toys" to practical productivity tools. The project agency‑agents surged to the top with a weekly increase of 10,976 stars. It simply defines 140 AI expert roles as Markdown files that can be imported directly into editors such as Claude Code, Cursor, and GitHub Copilot.

Developers are realizing that a single agent’s capabilities are limited, and collaborative multi‑agent workflows are the way forward. This is not an isolated case—over 80% of the Top 10 projects this week are AI‑agent related, indicating the emergence of a complete technical stack.

1. agency‑agents – AI Expert Team Framework

Weekly stars : +10,976

Not a framework but a collection of Markdown files, each describing an expert agent (frontend engineer, product manager, UI designer, community operator, etc.). It is plug‑and‑play and can be configured for Claude Code, Cursor, or GitHub Copilot.

Core value : Provides a "role definition standard" for multi‑agent collaboration, giving each agent a clear persona, workflow, and deliverable format.

Applicable scenarios : Independent developers needing multi‑role cooperation and small teams lacking certain specialist positions.

2. codebase-memory-mcp – Code Knowledge Graph

Weekly stars : +9,517

Enables AI coding assistants to have "memory" by indexing an entire codebase into a persistent knowledge graph, supporting 158 programming languages with sub‑millisecond queries and reducing token consumption by 99%.

Core breakthrough : Solves the pain point of LLMs re‑parsing code on every request. Indexing the Linux kernel takes only three minutes; subsequent queries run in milliseconds.

Applicable scenarios : Understanding large legacy codebases, dependency analysis before refactoring, and onboarding new team members quickly.

3. strix – AI Penetration‑Testing Tool

Weekly stars : +9,362

Runs your code in a Docker sandbox, dynamically attacks it, discovers vulnerabilities, and validates them with PoC scripts, automatically covering OWASP Top 10 issues.

Core innovation : Uses dynamic attack simulation instead of static rule matching, providing reproducible verification scripts and eliminating false‑positive noise.

Applicable scenarios : Security testing before release and supplementing understaffed security teams.

4. OpenMontage – Proxy‑Based Video Production System

Weekly stars : +8,447

Integrates 12 video‑production pipelines, 52 professional tools, and over 500 Agent skills. Users describe requirements in natural language; the Agent automatically gathers assets, writes scripts, records voice‑overs, edits, adds subtitles and effects.

Core value : Turns an AI coding assistant into a full‑featured video‑production studio.

Applicable scenarios : Content creators needing rapid video output and educational institutions producing batch courses.

5. ai‑berkshire – AI‑Powered Value‑Investing Framework

Weekly stars : +5,984

Systematizes the methodologies of Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu. Its key innovation is parallel multi‑agent research, launching several agents simultaneously to analyze a company from financial, moat, management, and valuation perspectives.

Applicable scenarios : Deep‑dive value‑investor research and learning diverse role‑based thinking.

6. SimpleX Chat – Account‑Free Private Messaging

Weekly stars : +4,630

First messaging network that uses no user identifiers—no usernames, phone numbers, or IDs—so the server never knows who is communicating with whom.

Core innovation : Protects even metadata via a simplex message queue and double‑ratchet encryption.

Applicable scenarios : Extremely high‑privacy communications and transmission of sensitive information.

7. video‑use – Programming Agent Video Editing

Weekly stars : +4,174

Users describe editing intents in natural language; the Agent calls FFmpeg to perform the work, removing the need to remember complex command‑line parameters.

Core value : Transforms video editing from "manual operations" to "intent expression".

Applicable scenarios : Automated video‑editing workflows and batch video processing.

8. OmniRoute – Free AI Gateway

Weekly stars : +4,133

Aggregates 236 AI providers (including 50+ free tiers) behind a single endpoint, offering automatic failover, load balancing, and pooled quota management.

Core value : Consolidates scattered free resources into a unified pool, maximizing the use of each platform’s free quota.

Applicable scenarios : Reducing AI‑tool usage costs and conducting multi‑model comparison tests.

9. daily_stock_analysis – LLM Stock‑Analysis System

Weekly stars : +3,842

Supports A‑shares, Hong Kong stocks, US stocks, and ETFs. It integrates market data, news, financial reports, and technical indicators, runs daily via GitHub Actions at zero cost, and automatically pushes analysis results.

Applicable scenarios : Personal investment decision support and automated data‑pipeline construction.

10. Orca – Parallel Agent Orchestration Platform

Weekly stars : +3,790

Allows CLI agents such as Claude Code and Codex to operate in parallel Git worktrees. Each agent gets an independent workspace, avoiding interference.

Core value : Runs five coding agents simultaneously in one window, boosting efficiency threefold.

Applicable scenarios : Large‑scale parallel development and simultaneous multi‑feature advancement.

Weekly Trend Summary

AI Agent Infrastructure Is Exploding

From role definition (agency‑agents) to memory layers (codebase‑memory‑mcp), orchestration platforms (Orca) to specialized applications (strix, video‑use), the AI‑agent ecosystem is forming a complete stack.

AI + Security Becomes a New Hotspot

Strix’s rapid growth signals the rise of AI‑driven penetration testing, which automates vulnerability discovery, PoC generation, and remediation, surpassing traditional static‑rule tools.

Parallel Multi‑Agent Workflows Are Now Standard

Orca’s parallel worktrees, agency‑agents’ expert‑team collaboration, and ai‑berkshire’s multi‑view research illustrate that parallel agents have become the default model for AI applications.

Open‑Source Replacements for Commercial SaaS Continue

Projects such as SimpleX Chat (vs. Signal), OmniRoute (aggregating paid services), and video‑use (vs. conventional editing software) show the community’s drive to challenge commercial monopolies with innovative open‑source alternatives.