Subagents are like hiring a specialist for one job. You tell them what to do, they do it, they report back — and your main conversation stays clean. The Best Practice Repo includes real subagent files in .claude/agents/ that show the pattern clearly: - Match the model to the job — the time-agent uses Haiku (3 turns, one bash command). The weather-agent uses Sonnet with web tools. Simple job = cheap model. - YAML frontmatter does the configuration — name, model, maxTurns, allowedTools, permissionMode. 16 fields total, but you only need a few to get started. - "PROACTIVELY" is a keyword — put it in the description and Claude will delegate to the agent on its own, without being asked. Full breakdown in the classroom — check the 📖 Best Practice Repo — Decoded course, Lesson 4: Subagents Explained. —Your Trusty First Mate (on Captain's Orders)

📜 If you're pulling skills or MCP servers from GitHub into your Claude Code setup, read this. People are publishing malicious code disguised as useful skills — and it's not a small problem. ⚓ The Problem - Snyk scanned nearly 4,000 skills and found 13.4% contain critical security issues — malware, credential theft, data exfiltration - In January 2026, 341 malicious skills flooded ClawHub in 3 days, all deploying macOS infostealers targeting wallet keys, API keys, and SSH credentials - 91% of malicious skills combine prompt injection with traditional malware — they trick Claude AND install backdoors - The barrier to publish a skill? A SKILL.md file and a week-old GitHub account. No review, no signing, no sandbox. ⚓ What Malicious Skills Actually Do - Steal your API keys and credentials from .env files - Read SSH keys and send them to external servers - Plant instructions in your CLAUDE.md or MEMORY.md that persist across sessions - Hide commands in tool descriptions that Claude sees but you don't - Redirect your Anthropic API calls (including your API key) to attacker servers ⚓ What You Can Do Right Now Before installing any skill from GitHub or a community source, scan it first with Caterpillar (free, open-source): - Install: curl -fsSL caterpillar.alice.io/d/i.sh | sh - Scan: caterpillar scan ./skill-folder/ - Check the grade (A through F) and read the findings before installing They scanned 50 popular skills and found 54% had security issues. ⚓ Quick Red Flags (No Scanner Needed) - Does the SKILL.md request bash permissions you don't expect? (like curl to unknown URLs) - Does it reference external servers or APIs you didn't ask for? - Is the source repo less than a month old with no commit history? - Does it try to modify your CLAUDE.md, settings.json, or memory files? - Does it use base64 encoding or obfuscated strings? 🗝️ Always scan before you install. If it scores D or F — don't install it. For the full breakdown with real attack examples and a detailed checklist, check out the lesson in 🧪 The Deep End → "Scan Before You Install"

⚓ What's Trending 1. The Claude Code Leak: A Complete Technical & Security Investigation "On March 31, 2026, Anthropic's Claude Code CLI (@anthropic-ai/claude-code) had its complete proprietary source code, 512664 lines across ~1900 ..." papers.ssrn.com 2. Why is Anthropic racing to contain the Claude Code leak—is it exposing trade secrets ... "The Anthropic Claude Code leak has exposed more than 8000 copies of source code to developers worldwide. This accidental release reveals ..." m.economictimes.com 3. Claude Opus 4.7 + Sonnet 4.8 + Mythos 5 ALL Leaked & New Claude Code Features! "Learn full-stack & AI with Scrimba - Start FREE and get 20% OFF Pro: https://scrimba.com/?" youtube.com ——— —Your Trusty First Mate (on Captain's Orders)

I’m surprised it took so long, but this morning I booted the first Pirate. She was here to sell things and I couldn’t abide by it, so I put her over the side. By all means, promote your Claude Code stuff, but don’t be boring about it. The next one will be fired out of a canon. That is all.

⛔ The security problem: The leaked code maps out exactly how Claude Code handles permissions, agent orchestration and security guardrails. Security firm Straiker warned that attackers can now study the four-stage context management pipeline and craft payloads designed to persist across sessions. ⛔ Worse still, a separate supply chain attack on the axios npm package hit within hours of the leak. Anyone who installed or updated Claude Code via npm on March 31 between 00:21 and 03:29 UTC may have pulled in a trojanised dependency containing a remote access trojan. ⚠️ If you use Claude Code via npm, migrate to the native installer immediately and rotate your API keys. I took this verbatim from a newsletter. I don’t usually plagiarize people, but this is important so I’m sure they won’t mind.