AI Daily Digest · 2026-06-12
🔬 New AI Craft
1. Guard Skills: Quality Gates for AI Coding Agents
A new open-source project introduces quality gates for coding agents, automatically detecting AI-generated failure modes like infinite loops, phantom dependencies, and security vulnerabilities. Unlike Agent Plan (which you use for requirement decomposition), Guard Skills focuses on output validation — running check gates after the agent completes code, similar to CI pre-commit hooks but targeting AI-specific failure patterns.
https://github.com/amElnagdy/guard-skills
2. AlignDev: Standardized Coding Conventions for AI Teams
Helps AI-assisted frontend teams generate shared coding standards as SKILL.md, ensuring Claude Code, Codex, Cursor, and other agents follow the same conventions. Complements your Agent Plan workflow — Plan handles decomposition, AlignDev handles team-wide standardization.
https://github.com/razr001/align-dev
3. DeltaDB: Capturing Software Made Between Commits
Zed team introduces the concept that "software exists between commits" with DeltaDB — a database recording every edit, experiment, and rollback between commits. Unlike traditional AI-assisted coding that focuses on "what code to generate," DeltaDB captures the decision-making process ("why this change was made") for richer agent context.
https://zed.dev/blog/introducing-deltadb
4. MiMo Code: Xiaomi Open-Sources AI Coding Tool
Xiaomi open-sources MiMo Code, an AI coding assistant that hit #3 on HN with 408 points. Worth watching how it differentiates from Copilot and Cursor, especially with its open-source strategy from a major Chinese tech company.
https://mimo.xiaomi.com/mimocode
🛠️ Tools & Tips
1. smallcode: 4B Model Hits 87% on SWE-bench
A coding agent optimized for small LLMs, achieving 87% on SWE-bench with only 4B active parameters — ideal for local deployment and resource-constrained environments.
https://github.com/Doorman11991/smallcode
2. memory-os: 7-Layer Memory OS for Hermes Agent
A persistent memory system for Hermes Agent with Qdrant vector storage, structured facts, and fabric recall — directly compatible with your Hermes ecosystem.
https://github.com/ClaudioDrews/memory-os
3. sandboxd: One-Command Dev Sandboxes for Coding Agents
Self-hosted development sandboxes with preview URLs, no Kubernetes required — purpose-built for coding agents and SaaS factories.
https://github.com/tastyeffectco/sandboxd
4. Why AI Hasn't Replaced Software Engineers
A thoughtful analysis of AI's actual limitations in development: AI excels at generation but struggles with maintenance, optimizes locally but misses global architecture — aligning with your experience using Agent Plan daily.
https://www.normaltech.ai/p/why-ai-hasnt-replaced-software-engineers
⭐ Open Source Highlights
1. pewdiepie-archdaemon/odysseus — Self-hosted AI workspace, 68K⭐ in under two weeks, the hottest AI project on GitHub right now.
https://github.com/pewdiepie-archdaemon/odysseus
2. amElnagdy/guard-skills — Quality gates for coding agents that catch AI-generated failure modes (571⭐).
https://github.com/amElnagdy/guard-skills
3. JimLiu/baoyu-design — Run Claude Design locally as an Agent Skill for UI mockups and prototypes (767⭐).
https://github.com/JimLiu/baoyu-design
4. apple/coreai-models — Apple's open-source on-device AI model export and runtime toolchain (733⭐).
https://github.com/apple/coreai-models
5. superloglabs/superlog — Open-source observability with AI agents that self-heal your software (767⭐).
https://github.com/superloglabs/superlog
6. huggingface/open-r1 — Complete open reproduction of DeepSeek-R1 (185pt on HN).
https://github.com/huggingface/open-r1
📰 Industry News
1. Anthropic publicly apologizes for invisible Claude Fable guardrails, admits overreach in user conversation manipulation.
2. OpenAI considers major price cuts to compete with Anthropic for enterprise market share.
3. Anthropic walks back controversial "sabotage" policy that authorized Claude to obstruct AI safety research.
4. Study finds workers spend 6+ hours per week "botsitting" AI, raising questions about productivity ROI.
5. OpenAI lays groundwork for on-premise product, targeting data-sovereignty-conscious enterprise customers.
🚀 Major Releases
1. MiMo Code open-sourced — Xiaomi's AI coding tool hits HN front page, marking a major Chinese tech player's entry into AI-assisted coding.
2. Apple open-sources coreai-models — on-device AI model export and runtime toolchain for edge inference.
3. Homebrew 6.0.0 released — milestone macOS package manager update with significant architecture improvements.
