Dev Digest — May 13, 2026

🔥 HOT RELEASES

Cline SDK — Open Source Agent Runtime Cline rebuilt their entire agent harness from scratch and open-sourced it. The SDK includes plugins, MCP integrations, cron jobs, subagents, checkpoints, and hub-backed sessions. 74.2% on Terminal Bench with Opus 4.7 — top of the leaderboard. This is the shift from "AI coding app" to programmable agent infrastructure. 🔗 X: https://x.com/cline/status/2054580767779700775 💻 GitHub: https://github.com/cline/cline/tree/main/sdk 📦 npm i @cline/sdk

OpenAI Daybreak — AI Cyber Defense Platform OpenAI launched Daybreak, combining GPT-5.5-Cyber and Codex Security for automated vulnerability detection and patch validation. Turns hours-long security analyses into minutes. Defenders and attackers both running AI stacks now. 🔗 X: https://x.com/RoundtableSpace/status/2054264090600808714 🌐 https://openai.com/daybreak/

Claude Agent SDK Credits (June 15) Anthropic announced dedicated monthly Agent SDK credits for paid plans, separate from chat limits. Covers claude -p, scripts, GitHub Actions, and third-party apps (OpenClaw, Conductor, etc). Pro gets base credits, Max/Team get $100-200+. 🔗 X: https://x.com/alexalbert__/status/2054613082589298899

Google Official Agent Skills Google dropped 13 official skills following the Agent Skills standard — Gemini API, BigQuery, Cloud Run, Firebase, GKE, and more. Compatible with Claude Code, Cursor, Copilot, Gemini CLI. The beginning of a real cross-agent ecosystem. 🔗 X: https://x.com/soulcodesmith/status/2054443123091358127 💻 GitHub: https://github.com/google/skills

eslint-rspack-plugin v5.0.0 Rewritten in TypeScript, pure ESM, install size reduced by 91%, drops ESLint v8, flat config by default. 🔗 X: https://x.com/jait_chen/status/2054492378627789227 💻 GitHub: https://github.com/rstackjs/eslint-rspack-plugin/releases/tag/v5.0.0

Claude Code 2.1.140 + Agent View New terminal dashboard for managing multiple concurrent AI coding sessions. Dispatch background sessions, monitor progress, reply inline. Plus: improved subagent matching, enterprise endpoint fixes. 🔗 X: https://x.com/TechOclockOff/status/2054303227919679907 💻 Changelog: https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md#21140

🧪 INTERESTING REPOS

InsForge — Context Engineering for Claude Code Cut Claude Code token usage by 3x (10.4M → 3.7M tokens, $9.21 → $2.81) by exposing auth, postgres, storage schemas via MCP. The agent reads instead of guessing. 🔗 X: https://x.com/DailyDoseOfDS_/status/2054494541470969969 💻 GitHub: https://github.com/InsForge/InsForge

agentmemory — Persistent Memory for AI Coding Agents 6000+ stars. Replaces static CLAUDE.md with retrieval-based memory (BM25 + vector + knowledge graph). 92% fewer context tokens with same recall. Works with anything that speaks MCP. 🔗 X: https://x.com/ghumare64/status/2054470602132492718 💻 GitHub: https://github.com/rohitg00/agentmemory

Orca — Agent Progress Viewer Track /goal progress across Claude Code, Codex, OpenCode. Open source agent view dashboard. 🔗 X: https://x.com/JinjingLiang/status/2054356206693785959 💻 GitHub: https://github.com/stablyai/orca

agent-of-empires — Multi-Agent TUI Command Center TUI/web dashboard for juggling Claude Code, Codex, Gemini, Copilot with tmux and worktrees. See every agent run, isolate worktrees, compare outputs. 🔗 X: https://x.com/alexmorris10x/status/2054615167590154704 💻 GitHub: https://github.com/njbrake/agent-of-empires

AGenUI — Alibaba's A2UI Framework End-to-cloud agent-to-UI framework for iOS, Android & HarmonyOS. Open-sourced by Qwen & Amap. 🔗 X: https://x.com/Alibaba_AMAP/status/2054520076100120692 💻 GitHub: https://github.com/AGenUI/AGenUI

NetWatch — Rust Network Diagnostics TUI Real-time network diagnostics in your terminal, written in Rust. 🔗 X: https://x.com/lcheylus/status/2054277688634855918 💻 GitHub: https://github.com/matthart1983/netwatch

OpenSign — Open Source DocuSign Alternative Create, fill, and sign digital documents. Trending. 🔗 X: https://x.com/tom_doerr/status/2054515673955528745 💻 GitHub: https://github.com/OpenSignLabs/OpenSign

CloakBrowser — Stealth Chromium Passes every bot detection test. Drop-in Playwright replacement with source-level fingerprint patches. 30/30 tests passed. 🔗 X: https://x.com/packetrat_/status/2054509729460707685 💻 GitHub: https://github.com/CloakHQ/CloakBrowser

AWS AI-DLC Workflows AWS open-sourced an SDLC methodology for AI coding agents. Rule pack that works with Claude Code, Cursor, Copilot, Codex, Kiro. 🔗 X: https://x.com/aiwire_x/status/2054520912775729618 💻 GitHub: https://github.com/awslabs/aidlc-workflows

cc-plugin-codex — Claude in Codex Sendbird's open-source plugin that runs inside Codex, lets you use Claude models for review, rescue, and tracked background workflows. 🔗 X: https://x.com/km_tr_/status/2054621464490553847 💻 GitHub: https://github.com/sendbird/cc-plugin-codex

🎥 WORTH WATCHING

Anders Hejlsberg on The Pragmatic Engineer Podcast The creator of Turbo Pascal, Delphi, C#, and TypeScript on Gergely Orosz's podcast. "This was such an interesting one!" 🔗 X: https://x.com/GergelyOrosz/status/2054554814613225716

Cursor's Head of AI Engineering — How They Build Coding Agents 14-minute talk on building coding agents powered by Claude. 30%+ of Cursor's internal merged PRs are now shipped by agents. Each agent gets its own VM, writes code, opens a browser. 🔗 X: https://x.com/0xCodez/status/2054611522530283636

MCP vs Agent Skills vs AGENTS.md Explainer Hands-on guide breaking down when to use each standard. 🔗 X: https://x.com/Pavan_Belagatti/status/2054568571485745183 🔗 Video: https://youtu.be/w65tyLWjGqA

💡 TECHNIQUES & IDEAS

Context Engineering > Prompt Engineering InsForge showed 3x token savings by exposing schemas to agents via MCP instead of letting them guess. Karpathy's reminder: most AI coding costs go to context the model didn't need. "Before you grep the repo, ask what the agent actually needs." 🔗 X: https://x.com/ianneo_ai/status/2054597537957650462

From Prompting to Persistent Jobs The shift from "chat with AI" to "supervise work until acceptance criteria pass." Coding agents becoming task runners with goals, checkpoints, and recovery loops. 🔗 X: https://x.com/nullhypeai/status/2054543931300560948

Google Cloud AI Director's Agent Workflow System Full development lifecycle broken into explicit stages: DEFINE → PLAN → BUILD → VERIFY → REVIEW → SHIP. 22 reusable skills, 7 slash commands, multiple agent personas. Free. 🔗 X: https://x.com/Chestu_eth/status/2054515127823556851

Token Tracker for Claude Code / Codex Local dashboard showing 5h/7d quota, context remaining, model, token usage, session cost. CLI + reports. MIT open-source. 🔗 X: https://x.com/kaanuzdogan/status/2054476814123159847

🔮 EMERGING TRENDS

Agent Infrastructure Race Cline SDK, Google Skills, AWS AI-DLC, Cloudflare Agent Setup — everyone is building the platform layer. The question isn't "which AI writes code" but "which agent infrastructure wins." Coding agents are becoming operating systems.

Multi-Agent Orchestration Goes Mainstream Agent View in Claude Code, agent-of-empires TUI, Cline subagents, ruflo swarm platform — managing multiple agents simultaneously is the new UX challenge. "Agentic coding gets powerful when orchestration becomes visible."

Agent Skills as Portable Assets Google adopting the standard, Cline SDK supporting npx skills add, AWS publishing methodology packs — skills are becoming the unit of agent capability, portable across tools.

Security Agents OpenAI Daybreak, Codex Security — AI moving from writing code to protecting it. Red team/blue team self-play at scale. "The product surface moved from 'AI writes code' to 'AI inherits the engineering organization.'"

The Great Claude Code Migration Devs leaving Cursor/Copilot for Claude Code CLI. "No real reason to pay for Cursor anymore." NinjaOne moving 700 engineers from Copilot to Claude Code. Meanwhile, Anthropic splitting SDK usage from chat limits — acknowledging agents are the future.

Compiled from X/Twitter on May 13, 2026

Full digest: https://digest.99.cooking