Hello again Midnight community,
Following up on my initial post about Midnight Agent Skills ( GitHub - mzf11125/midnight_agent_skills · GitHub ), I want to share a massive update (v2.0.1) we just rolled out to keep AI assistants perfectly in sync with the bleeding edge of the Midnight ecosystem.
As the network rapidly evolves, AI models quickly fall behind. Over the past few days, I’ve completely overhauled the skill modules to align with the latest official documentation (docs.midnight.network/llms.txt), GitHub repositories, and April 2026 developments.
What’s new in this update:
- Full Migration to Preprod
All deprecated testnet-02 and preview references have been completely purged from the repository. All 4 skills, the Python helper scripts, and the 18 deep-dive reference documents now natively target the active Preprod environment (including the latest v4.0.0 Indexer GraphQL endpoints). - Programmatic DUST Generation
Since DUST is no longer directly available from the faucet on Preprod, AI agents were getting confused. The midnight-concepts and midnight-api skills now explicitly teach agents the 4-phase DUST lifecycle and how to use the official midnight-dust-generator to register and generate DUST programmatically. - Midnight MCP Integration (Fixing Code Hallucinations)
This is a game changer for Compact development. The midnight-compact skill now natively instructs agents on how to use the Midnight MCP server ( GitHub - Olanetsoft/midnight-mcp: Midnight MCP server giving AI assistants access to Midnight blockchain — search contracts, analyze code, explore docs · GitHub ). Instead of guessing syntax, AI assistants (like Claude Code and Cursor) can now leverage 29 custom tools to physically compile, analyze, and validate Compact contracts before presenting them to you. - CI/CD & Governance
- GitHub Actions: Agents now know how to set up automated testing pipelines using the official setup-compact-action.
- MIPs: Added the formal Midnight Improvement Proposals (MIPs) process, so agents understand how to help you draft standard interfaces (Core, Standards, Networking, Governance).
-
Real-world DApp References
Agents are now pointed directly to official template implementations (like example-bboard, example-zkloan, and OpenZeppelin contracts) to use as architectural scaffolds rather than inventing patterns from scratch.If you are using Claude Code, Cursor, or other AI assistants for your Midnight development, installing these updated skills will massively reduce hallucination and boilerplate setup time.
You can check out the updated v2.0 release here:
GitHub - mzf11125/midnight_agent_skills · GitHub
As always, I’d love to hear how this is working for your local development and if there are any specific workflows or DApp patterns you’d like to see added to the knowledge layer!