- Oct 22: Anthropic hands the Agent Skills standard — spec, name, logo, GitHub org, and agentskills.io — to the Linux Foundation's Agentic AI Foundation (AAIF), the same neutral home it set up for MCP.
- 1 million skills published and nearly 280 million installed to date; more than 70 agent platforms integrated, including ChatGPT, Cursor, GitHub Copilot, and VS Code.
- Engineers from Anthropic, Google, OpenAI, and others will co-maintain as individuals; David Soria Parra and Mahesh Murag are the Lead Maintainers.
- Roadmap: versioned releases, a formal proposal process, and security guidance for authors and installers of skills.
- cloudflare/security-audit-skill: 40.0k stars, +2,979 today — up roughly 14k in two weeks since its Oct 8 cameo at 26k.
- Six-phase audit: recon, coverage-led hunting with isolated sub-agents, validation by a fresh verifier that tries to disprove each lead, structured findings.json, independent re-verification, report.
- Ships attack-class prompt packs for memory safety, LLM apps, cloud config, and supply chain; needs an OS-sandbox or leads stay as needs_validation.
- MIT. Seeded Cloudflare's internal vulnerability discovery harness; installs via `npx skills add cloudflare/security-audit-skill`.
- Graphify-Labs/graphify at 133.2k stars (+1,408 today) — tree-sitter AST parsing locally, no embeddings, no vector index.
- Edges tagged EXTRACTED or INFERRED; outputs graph.html, GRAPH_REPORT.md, and graph.json for queries, shortest paths, and concept explanations.
- 0.497 recall@10 on LOCOMO with zero LLM credits for the graph build; semantic pass over docs, PDFs, and video is optional.
- Ships as an Agent Skill for Claude Code, Cursor, Codex, Gemini CLI, Copilot CLI, and ~20 other harnesses; Apache-2.0 and MIT.
- HKUDS/CLI-Anything at 56.3k stars (+3,275 this week) — pitched as 'making ALL software agent-native' via auto-generated CLIs on top of desktop apps.
- Covers GIMP, Blender, LibreOffice, and Audacity today; CLIs expose JSON output, a REPL, and discoverable --help interfaces.
- 7-phase generator (analyze, design, implement, plan tests, write tests, document, publish) builds new harnesses from any codebase.
- Companion CLI-Hub registry for installing community CLIs; Apache-2.0 and ships as a Claude Code / Cursor / Codex / Copilot plugin.
- anthropics/knowledge-work-plugins at 31.8k stars (+303 today) — bundles skills, connectors, slash commands, and sub-agents for whole job functions.
- 11 role packs: productivity, sales, customer-support, product-management, marketing, legal, finance, data, enterprise-search, bio-research, plus a plugin-management meta-pack.
- Everything is Markdown + JSON, no code or infra; MCP wiring to Slack, Notion, Jira, HubSpot, Snowflake, Databricks, PubMed, Benchling, and more.
- Installs in Claude Code via `claude plugin install sales@knowledge-work-plugins`; Apache-2.0.
- ethanplusai/harvey: 2.2k stars (+1,300 this week), MIT — autonomous cold-email + prospect-research loop powered by Claude Code or Codex.
- No API bill — uses your Claude Pro/Max or ChatGPT subscription; prospect data comes from free OpenStreetMap by default.
- Rule-based decision loop every ~15 min: prioritizes replies, writes campaigns, prospects; every outbound email waits for approval unless you flip autopilot.
- Latest tag in a wider pattern of agentic apps piggybacking consumer AI plans to dodge metered inference costs.
01
Anthropic Donates Agent Skills Standard to the Agentic AI Foundation
frameworks anthropic.com
02
Cloudflare's security-audit-skill Rockets Past 40K Stars on GitHub Trending
open-source github.com
03
Graphify Turns Codebases Into Queryable Knowledge Graphs Across 40 Languages
tools github.com
04
CLI-Anything Hits 56K Stars Wrapping GIMP, Blender, and LibreOffice for Agents
open-source github.com
05
Anthropic Open-Sources 11 Role-Specific Plugin Packs for Claude Cowork
tools github.com
06
Harvey Open-Sources an Autonomous Sales Agent That Rides Your Claude or ChatGPT Subscription
tools github.com