Agent skills checklist shared

A fast-growing open-source project is telling teams to treat artificial intelligence coding agents like junior engineers with checklists, not magic autocomplete. (github.com) The repository, `agent-skills`, is published by Addy Osmani on GitHub and describes itself as “production-grade engineering skills for AI coding agents.” GitHub showed about 10,300 stars and more than 1,100 forks when it was crawled in the past few days. (github.com ) Its core idea is simple: an agent should move through the same software lifecycle a human team uses — define, plan, build, verify, review and ship. The project maps that flow to seven slash commands, including `/spec`, `/plan`, `/build`, `/test`, `/review`, `/code-simplify` and `/ship`. (github.com) The repository says those commands are only the entry points. Underneath them are 19 named skills, including spec-driven development, test-driven development, code review and quality, security and hardening, performance optimization, documentation and architectural decision records, and continuous integration and deployment automation. (github.com) That framing shifts the conversation around coding agents away from picking a model and toward setting rules for how work gets done. The README says agents “default to the shortest path,” which in practice can mean skipping specifications, tests and security reviews unless teams make those steps explicit. (github.com) The project also packages those rules in plain Markdown files rather than a model-specific format. Its getting-started guide says any coding agent that accepts instruction files can use the skills, and it includes setup guides for Claude Code, Gemini Command Line Interface, Cursor, Windsurf, GitHub Copilot and Codex-style tools. (github.com, github.com) One of the more concrete ideas is that the agent should prove work in small slices instead of delivering a large batch at the end. The README ties `/plan` to “small, atomic tasks,” `/build` to “one slice at a time,” and `/test` to “tests are proof.” (github.com) The skills are written as workflows with verification gates and what the project calls “anti-rationalization tables,” meant to stop an agent from talking its way around missing evidence. The documentation says a skill is not a reference note but a step-by-step process with exit criteria. (github.com, github.com) Shubham Saboo, whose post helped circulate the checklist, describes his GitHub work as practical tutorials and starter kits for artificial intelligence agents, retrieval-augmented generation and large language model applications. His repositories page also shows a small `agentskills` repository updated in recent weeks. (github.com, github.com) The larger point is less about one repository than about how teams are starting to manage agent output. In this view, reliable agent coding comes from specifications, tests, reviews and deployment gates that can be measured and repeated, not only from swapping in a newer model. (github.com, github.com)