Snack Pack #1 · Oct 16, 2026
Snack Pack #1: Your agents got a babysitter
Preview — Pack #1 is on the site first. The Thursday email goes out when we hit send.
Week of Oct 9–16, 2026 · 7 snacks · ~6 min
Agents learned to babysit themselves this week. Grab a coffee — they’ll keep chewing.
Nutrition Facts
Serving: 1 Snack Pack · ~6 min
Counts from this issue · 15 cards · Hype: 0 g.
Hey — welcome to Snack Pack #1. Same deal every Thursday: only what actually shipped in the AI coding-agent tools you already use, each bite with a source link. No funding fluff. No “AI will change everything.” Just the version bumps that might change your Friday.
- nice to know
- worth a look
- act this week
Agent Skills: recipe cards for your AI
Agent Skills are recipe cards for your AI. A tiny name+description loads at startup, and the full instructions only show up when a task matches (Claude blog: Skills). Try this: skim one skill folder in anthropics/skills before you paste another 2,000-word essay into CLAUDE.md. New here?
Source: claude.com · skills · github.com · anthropics/skills
Copilot will now fix your security warnings for you
Copilot can now fix your security alerts and prove it. Assign a CodeQL alert and it patches the code, re-runs CodeQL to prove the alert is gone, then opens a draft PR. Up to 25 alerts per batch. Now GA.
Default-branch alerts only: an enterprise admin has to turn on the “Agentic autofix” policy, and each run spends Code Security AI Credits + Actions minutes. You can also start it via the REST API (copilot-swe-agent[bot]).
Who cares Security folks staring at a CodeQL mountain on GHEC / GHAS.
Source: github.blog · changelog
HydraFusion stops making you pick a model (and adds a “Max” mode)
HydraFusion gets a bigger model pool and a Max tier. Its model pool now includes Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. Cascade gets a judging step; Critique can tool-snoop the workspace with parallel critics. New HydraFusion Max is the “go deep” button.
Lives in VS Code, the Copilot app, and Copilot CLI.
Who cares Copilot Pro+ / Business folks tired of the model roulette wheel.
Source: github.blog · changelog
Claude Code: idle chats now compact before the cache expires
Walk away from a long chat and Claude Code tidies it up. After ~54 idle minutes, a 200K+ token chat compacts before the prompt cache expires, so your next message doesn’t re-chew the whole transcript. Tune the threshold with CLAUDE_CODE_IDLE_COMPACT_MIN_TOKENS, or opt out with "idleCompaction": false.
Who cares Anyone with marathon Claude Code sessions — or a token bill that makes them wince.
Copilot CLI: hooks that don’t hang the babysitter
Autopilot stops freezing on questions nobody is there to answer.
- v1.0.96:
/add-dirgrants sandbox access for the session; interactive settings nudge you to mask likely secrets. - v1.0.97: agent-scoped declarative hooks; Autopilot now denies a preToolUse “ask” instead of freezing on a dialog nobody is watching.
- v1.0.98: Gemini 4 Argon support;
/askexpands with Ctrl+E, copies with Ctrl+Y.
Who cares Terminal Copilot users running Autopilot / unattended loops.
Cursor Origin opens the snack aisle
Cursor’s Origin now plugs into the apps you already use. That’s 20+ apps: Cloudflare, Vercel, Buildkite, CircleCI, CodeRabbit, Greptile, Semgrep, Socket, Sentry, Linear, and friends. Public submissions welcome. Apps auth with app JWT + install tokens, get webhooks, write check runs / PR comments / reviews.
Who cares Dev-tool makers hunting distribution, and teams wiring CI/review bots into Origin.
Source: cursor.com · changelog
Cline swapped your default model
Who cares Cline users who never pinned a model. Check the cost dashboard before you blame the agent.
Source: github.com · v4.1.24 · desktop-v0.0.46
OpenCode 1.19: the on-ramp to OpenCode 2 (and auto-update is off)
opencode upgrade --major (or /update in the TUI) walks you there in v1.19.0, so upgrade on purpose. Also allows OpenAI’s ultrafast service tier.
Who cares OpenCode folks. Deliberate upgrades > surprise majors.
Source: github.com · v1.19.0
Agent Skills in plain English: recipe cards for your AI
Tired of re-explaining the same steps to your AI every single chat? An Agent Skill is a recipe card it keeps in a folder (SKILL.md): only the name and blurb sit on the menu at startup (about 100 tokens), and the full recipe comes out when your task matches.
Skills teach your agent how to cook; new hands to grab your tools are MCP’s job.
~6 min read
Source: platform.claude.com · context windows · overview · claude.com · skills · modelcontextprotocol.io · intro · github.com · v2.1.299
pin your model before the defaults move
Try it now — pin your model before the defaults move:
- Open your agent settings (Cline, Copilot, Claude Code, Cursor — wherever you live).
- Explicitly pick the model you want this week so a quiet release cannot swap the chef under you.
- Recheck cost dashboards after any "default model" note in release notes — surprises usually show up on the bill first.
Claude Agent Skills Explained
Video Summary
- Skills are expertise Claude loads only when it needs it. Each one is a folder that kicks in automatically when a task matches.
- Only each skill's name and description load at startup; the full SKILL.md loads on demand.
- Compares skills with CLAUDE.md (project facts), MCP (connects to data) and subagents (fixed roles).
On screen: Mostly a code editor: skill .md files, CLAUDE.md, an mcp.json with Playwright, database-migrations.md (model: sonnet), and full-stack-features.md stacking skills.
After watching you can: Tell skills, CLAUDE.md, MCP servers and subagents apart, and know when to use each.
Short on time? Watch just:
Source: youtube.com · Watch
MCP reference servers 1.0
The classics grew up. Filesystem, memory, everything and sequential-thinking hit 1.0.0 on npm; fetch, git and time got fresh PyPI builds.
Source: github.com · v1.0.0
Anthropic claude-api skill
Meet Claude’s cheat sheet for the Messages API. It covers dynamic workflows, four quickstarts, and Haiku 5.5 as the current Haiku.
Source: github.com · skills/claude-api
GitHub MCP Server v2.0.0
Your agents can stop guessing what GitHub sends back. GitHub’s official server now ships structured outputs and output schemas for Code Mode agents (schemas need MCP 2026-07-28+ clients).
Source: github.com · v2.0.0
The agents got a babysitter
- Update now: Gemini CLI v0.64.0 tightens folder-trust in headless mode (v0.64.0 release).
- Devin Desktop 3.10.67 speaks a newer MCP revision and lets admins pin the extension marketplace.
- Codex CLI shipped v0.162.1 bugfix plus a pile of 0.163.0 alphas.
- No Aider release in the window.
Source: github.com · v0.64.0
That’s the pack. Pass a snack to a friend.