Guardrails for AI-assisted software development — a collection of agents, skills, and quality guidance that keep agentic coding tools productive and consistent.
agents/— Opinionated "craftsperson" agents for 16 language/framework stacks. Each agent enforces quality gates, TDD workflow, and language-idiomatic patterns. See agents/README.md.skills/— Reusable skills following the Agent Skills specification. See skills/README.md.conventions/— Application-level design conventions (CLI UX, and more to come) gathered as plain documentation before graduating into skills. See conventions/README.md.AGENTS.md— Baseline standards (versioning, frontmatter, file conventions) shared by all agents and skills in this repo.
AI coding agents are powerful but unconstrained — they'll happily skip tests, ignore linters, invent abstractions nobody asked for, and commit without running quality checks. These guidelines give agents a clear set of engineering principles to follow:
- Quality gates — tests, linting, formatting, and security scans must pass before work is done
- Simple Design Heuristics — tests pass, reveals intent, no knowledge duplication, minimal entities
- Functional core / imperative shell — pure logic at the center, I/O at the boundaries
- TDD workflow — red, green, refactor
- Documentation sync — docs stay aligned with code
Copy or symlink the agents and skills you need into your project's agentic tool configuration. Each agent is a standalone .md file; each skill is a self-contained directory with a SKILL.md.
For details on creating your own agents, see agents/MAKERS.md.
Stacey Vetzal — stacey@vetzal.com