🛠️ 9 Drops for the Operator Stack Today

Claude Code skills went from "cool pattern" to "fastest-moving category on GitHub" practically overnight. Today's drop has 9 repos worth opening: four are Claude Code skills from the awesome-claude-code list, two are curation indexes that save hours of searching, one is an open-source voice studio at 32k stars, and two are heavyweight anchors for the operator toolkit. One skill teaches your agent to run Xcode builds. Another adds Swift Concurrency expertise as a loadable module. A third gives Obsidian CLI access to any agent that can read Markdown. The category is maturing faster than most operators realize.

There's also real signal in the Signals section today. The Five Eyes intelligence agencies issued a rare joint statement warning that AI models capable of destabilizing governments are months away, not years. That's not hype from a think tank. That's signals agencies in five countries putting their name on it.

Nine repos, three Signals worth reading, one Playbook move you can steal before lunch.

[Skill] kepano/obsidian-skills, 36,781 stars. Agent skills for Obsidian. Teaches your agent to use the Obsidian CLI plus open formats: Markdown, Bases, JSON Canvas. If your knowledge base lives in Obsidian, this is the wire that connects it to Claude Code without copy-paste.

[Repo] jamiepine/voicebox, 32,091 stars. The open-source AI voice studio. Clone voices, dictate content, create audio from text. At 32k stars and trending today, this is the community's clearest signal on where open voice tooling is heading.

[Repo] bytedance/UI-TARS-desktop, 37,051 stars. Open-source multimodal agent stack for desktop. Connects frontier models to agent infrastructure so your agent sees and acts on a real UI. The operator play: autonomous QA and desktop task automation without writing a Playwright script from scratch.

[Skill] AvdLee/Swift-Concurrency-Agent-Skill, 1,559 stars. Loadable skill that adds expert Swift Concurrency guidance to any AI coding tool supporting the Agent Skills open format. Covers safe concurrency, performance optimization, and Swift 6 migration. Ship it once, load it in every iOS session.

[Skill] conorluddy/ios-simulator-skill, 1,100 stars. Claude Code skill that wraps xcodebuild so Claude can build, run, and interact with your iOS app in the simulator without burning context on verbose Xcode output. The gotcha it solves: raw xcodebuild logs are context-expensive and Claude loses the thread. This trims it.

[Skill] daymade/claude-code-skills, 1,207 stars. A production-ready skills marketplace for Claude Code. Think of it as a curated shelf of reusable abilities you install once and invoke with /name. Beats rolling your own from scratch when someone already did the hard part.

[Repo] rohitg00/awesome-claude-code-toolkit, 2,149 stars. The most comprehensive Claude Code index I've seen: 135 agents, 35 skills, 42 commands, 176+ plugins, 20 hooks, 15 rules, 7 templates, 14 MCP configs, 26 companion apps. If you are building your Claude Code setup from scratch, start here before building anything yourself.

[Repo] ai-for-developers/awesome-ai-coding-tools, 1,856 stars. Curated list of AI-powered coding tools across every major category. Useful when you need to audit what exists in a category before committing to one tool. Less Claude-specific than the toolkit above, wider surface area.

[Repo] run-llama/llama_index, 50,291 stars. LlamaIndex is the leading document agent and OCR platform. If your agent needs to reason over PDFs, contracts, or large document collections without a custom pipeline, this is the mature choice. The star count reflects it.

[MCP] github/github-mcp-server, 30,884 stars. GitHub's official MCP server. Wire it in and Claude can read repos, open issues, create PRs, and check CI status directly in-session. Add it to your project scope with a .mcp.json at the repo root and confirm with /mcp. The non-obvious move: use it to let Claude read its own issue queue and self-assign next steps.

[MCP] runapi-ai/mcp RunAPI's MCP server for model discovery, pricing lookup, task creation, and LLM chat. Useful when you are running multi-provider setups and want Claude to route tasks to the cheapest model that fits the job. Discovery and pricing in-session means no tab-switching to compare providers mid-build.

[Skill] numman-ali/n-skills, 999 stars. Curated plugin marketplace for AI agents. Works with Claude Code, Codex, and the open openskills format. Good secondary index if daymade/claude-code-skills above doesn't have what you need.

Want to run a setup like this yourself? aigent-OS ($49) →

Five Eyes agencies issued a rare joint warning: AI models capable of disrupting governments and major institutions are months away, not years. The statement names Anthropic's Fable model explicitly, citing Trump's decision to block foreign nationals from accessing it as context. This is not a blog post. This is Australia, the US, the UK, New Zealand, and Canada, officially on record. For operators building on frontier models, it is the clearest signal yet that access restrictions, compute controls, and export rules are getting tighter, not looser. (The Guardian)

OpenAI launched "Patch the Planet," a Daybreak initiative to find, validate, and fix open-source software vulnerabilities using GPT-5.5-Cyber and expert review. The operator consequence: if this works at scale, AI-assisted security audits become a baseline expectation, not a differentiator. Open-source maintainers get a free resource. Operators who ship on top of open-source should watch which CVEs get patched and how fast. (OpenAI Blog)

GLM-5.2, a new open-source model from China, is generating the kind of industry attention not seen since DeepSeek R1. The Economist and Business Insider both covered it the same week. America's lead over China may be at its smallest in over a year. For operators choosing a cost-efficient base model, GLM-5.2 is worth a benchmarking run before the next billing cycle. (Business Insider)

This week: hooks. Auto-run any command on a Claude Code event, without touching the prompt.

Hooks let you fire a shell command the moment Claude takes a specific action: before a tool runs, after a file edit, when a bash command matches a pattern. The payoff is ambient enforcement: linting that always runs, dangerous commands that never execute, and a log of every tool call for the sessions where you need an audit trail.

1. Create a hooks config in your project: add a hooks key to your .claude/settings.json (project scope) or ~/.claude.json (user scope). Each entry names the event (PreToolUse, PostToolUse, Stop) and a matcher, then a shell command to run. 2. To block a dangerous bash pattern, add a PreToolUse hook matching Bash with a command that inspects $CLAUDE_TOOL_INPUT and exits non-zero if it sees rm -rf / or a production DB connection string. Claude treats a non-zero exit as a veto. 3. To auto-lint on every file write, add a PostToolUse hook matching Write that runs your linter against $CLAUDE_TOOL_OUTPUT_PATH. Claude sees the lint output and can self-correct before you review.

You will know it worked when Claude tries to run a blocked command and the session surfaces your veto message instead of executing. The first time it stops itself, hooks earn their setup cost.

Claude Code docs: hooks

Move: Wire Obsidian as your agent's live knowledge base.

Skills repos from today's Drops make this a 10-minute setup. Here is the move:

1. Install kepano/obsidian-skills by dropping the skill file into your .claude/skills/ directory. 2. Point the skill at your Obsidian vault path in the skill config so Claude knows where to read and write notes. 3. In your CLAUDE.md, add one line: Use the obsidian skill to read project notes before starting any task and to write a summary note after completing one.

You will know it worked when Claude opens a task, reads the relevant note from your vault, and writes a completion summary you can actually search later. Your vault becomes a running operator log without any manual upkeep.

The issue you are reading was researched, drafted, and shipped by aigent-OS: the enhancement layer that turns a stock Claude Code session into the operator that runs this newsletter every day. Same Claude you already pay for, enhanced into a system that knows your project, routes work to a 9-specialist crew, and surfaces the right skill the moment a task needs it. Not a template, not a demo, a live daily operation. One install runs it inside your own Claude Code. Founding price $49 (then $79), 30-day money-back.

The skills category exploded quietly. No one lab announced it, no press release, just 30,000 developers starring voice studios and building Xcode wrappers and Obsidian connectors on a shared open format. That is what the ground looks like when the gatekeeping actually ends. Pick one repo from today's Drops, clone it before the day is out, and see what your setup looks like with one more loaded skill.

See you Wednesday.