AI coding assistants are only as good as the instructions you give them. Most repos give them nothing, so they guess: wrong test commands, no commit discipline, no idea what "done" means for your project.
I packaged the setup a senior engineer would write for your repo into a single drop-in kit: The Agentic Coding Kit.
What's inside (34 files)
- CLAUDE.md presets for 5 stacks - Python/FastAPI, TypeScript/Node, React/Next.js, Go, Rust. Scoped permissions, build/test commands, and conventions the agent follows from the first prompt.
- 6 Cursor rules (.mdc) - testing, git discipline, security review, refactor guardrails, docs, style.
- 5 Claude Code subagents - code reviewer, test writer, security auditor, refactorer, doc writer.
-
4 slash commands -
/review,/test,/secure,/ship. - 3 hook recipes with ready-to-use scripts - block force pushes, require tests before commit, auto-lint on save.
- CI workflow templates, pre-commit config, ADR templates, MIT license.
Why it works
Agents drift when the rules live in your head. A CLAUDE.md that says "run pytest before committing, never force-push, keep diffs under 400 lines" turns a chaotic assistant into a predictable one. The kit gives you that file, tuned per stack, plus the hooks that enforce it mechanically.
Unzip into your repo, pick the preset for your stack, done. Every file is plain Markdown, JSON, or shell - no lock-in, read everything before it runs.
Get The Agentic Coding Kit - $19, one-time
Free updates for all v1.x releases. If it saves you one afternoon of wrangling a runaway agent, it paid for itself.
Top comments (0)