← All Posts

Forkcast Weekly #17: ECC, ponytail, timesfm, academic-research-skills, hermes-agent, humanizer

Episode 17 covers 30 trending open-source projects from this week including a 250K-star agent harness optimization system with persistent memory and

Episode trending-weekly ID: 2026-09-06-trending-weekly-ep17 #forkcast#trending-weekly#ecc#ponytail#timesfm#academic-research-skills#hermes-agent#humanizer

Watch Episode

Episode Summary#

Episode 17 covers 30 trending open-source projects from this week including a 250K-star agent harness optimization system with persistent memory and security guardrails, a YAGNI-enforcing skill that cuts agent token waste by 65%, Google Research’s time-series foundation model that works across any domain, an academic research pipeline automating the full write-review-revise cycle, and a 242K-star persistent agent that grows with you. Also featured are an AI writing humanizer using 35 Wikipedia patterns, Chrome’s official DevTools MCP server for coding agents, a 64M-parameter LLM you can train in 2 hours for $3, and an open-source SEO alternative to Semrush.

Repositories Covered#

ECC#

Agent harness performance optimization system with skills, instincts, persistent memory, and security guardrails for Claude Code, Codex, OpenCode, and Cursor. 250K stars, 354 contributors, JavaScript 70.9%, Rust 20.9%. MIT license.

ponytail#

Makes your AI agent think like the laziest senior dev. Forces YAGNI principles, cuts unnecessary code, marks deliberate simplifications with grep-able ponytail comments. 128K stars. MIT license.

timesfm#

Google Research pretrained time-series foundation model. Handles univariate and multivariate forecasting with 9 quantile outputs across any domain. 31.4K stars. Apache 2.0.

academic-research-skills#

Full academic pipeline: research, write, review, revise, finalize. v3.21.2 works with Claude Fable 5.1 and GPT-6 Astra. 46.5K stars, 755 commits.

hermes-agent#

Persistent agent from Nous Research that grows with you. Local-first memory, cross-platform, develops working relationships over time. 242K stars, 32K commits. MIT license.

humanizer#

Agent skill that removes 35 known patterns of AI writing using Wikipedia’s Signs of AI writing checklist. Two-pass rewrite with voice matching. 43.6K stars. MIT license.

chrome-devtools-mcp#

Official Chrome DevTools MCP server for coding agents. Performance traces, console messages, network requests, heap snapshots, PWA automation. 51.1K stars. Apache 2.0.

minimind#

Train a 64M-parameter LLM from scratch in 2 hours for $3 on a single RTX 3090. Full pipeline: MoE, pretrain, SFT, LoRA, RLHF, DPO, PPO, GRPO, Tool Use. 58.9K stars. Apache 2.0.

caveman#

Claude Code skill that cuts 65% of tokens by making your agent talk like caveman. Why use many token when few token do trick. 104K stars.

pdf-inspector#

Fast Rust library for PDF inspection, classification, and text extraction. Detects scanned vs text-based PDFs for smart routing. 18.9K stars.

ODS#

Turns your PC, Mac, or Linux box into a full AI server. LLM inference, chat UI, voice, agents, workflows, RAG, and image generation. 6.2K stars, 3,266 commits. Apache 2.0.

open-seo#

Open-source alternative to Semrush and Ahrefs. Keyword research, backlink analysis, site audits, rank tracking. MCP servers for Claude Code and Codex. 17.3K stars. MIT license.

skills#

Anthropic’s official public repository for Agent Skills. Canonical, tested, maintained skill sets. 175K stars.

magnitude#

Release pipeline built for coding agents. Handles the gap between agent-written code and production deployments. 3.3K stars. Trending 4 times this week.

video-use#

Edit videos with coding agents. Describe cuts, transitions, and effects in natural language. From the browser-use team. 24.2K stars.

reverse-skill#

AI-powered security skill router with on-demand toolchain bootstrapping and self-evolving knowledge base. 34.7K stars. Supports Claude Code, Kiro, Cursor, Cline.

atlas#

Source control for agents. Tracks agent modifications separately from human changes with clean audit trails and easy rollback. 3.2K stars.

exploitarium#

Single archive of public exploit PoCs and vulnerability research writeups. 4.8K stars. For education and security research.

reclip#

Download videos from almost any website. Lightweight, self-hosted media downloader with a clean web UI. 8.7K stars.

patent-disclosure-skill#

Chinese patent disclosure skill: mining patent points, writing invention/utility/design disclosures, policy monitoring, and review assistance. 7.5K stars.

TypeWords#

Practice English typing, one strike, one step forward. Combines language learning with typing practice for real fluency. 9.6K stars.

user-scanner#

2-in-1 Email and Username OSINT suite. Analyzes 465+ scan vectors for security research and digital footprinting. 4.7K stars.

awesome-design-md#

Collection of DESIGN.md files from popular brand design systems. Drop one into your project and coding agents generate matching UIs. 114K stars.

FckSignups#

Curated list of open-source, in-browser tools that require zero signups. Every tool runs entirely in your browser. 3K stars.

portless#

Replace port numbers with stable, named local URLs. From Vercel Labs. For humans and agents. 12.2K stars.

diagram-design#

38 editorial diagram types for Claude Code, Codex, and Pi. Self-contained HTML plus SVG. No shadows, no Mermaid slop. 31.9K stars.

skills#

Curated, production-tested skill set for AI agents from HumanLayer. 2.8K stars. Trending twice this week.

Sequoia-X#

A-share automatic stock selection system with multiple technical pattern scanning. Runs after market close and pushes to Feishu. 6.8K stars.

everything-claude-code#

Claude Code toolkit: agents, commands, skills, rules, and hooks for productive AI-assisted development. 2.4K stars.

humanlayer#

Human-in-the-loop approval layer for AI agents. Agent asks permission before destructive operations. 2.8K stars.

Watch#

Notes#

Transcript and notes will be added from Forkcast output artifacts.