style

A language-agnostic, single-source-of-truth coding-style harness that feeds my practices into any AI coding agent.

Install

git clone https://github.com/marquesds/style ~/.style
cd ~/.style
./install.sh                   # installs into all detected agents
./install.sh --agent claude    # only one agent
./install.sh --dry-run         # show what would change, write nothing

The installer writes only into a managed namespace inside each agent's config directory and never clobbers files outside it. Re-running is idempotent.

Uninstall

./uninstall.sh                 # remove harness from every detected agent
./uninstall.sh --agent cursor  # only one agent
./uninstall.sh --dry-run       # show what would be removed, write nothing

The uninstaller deletes only files carrying the <!-- style-harness:managed --> marker and strips the <!-- BEGIN style-harness --> … <!-- END style-harness --> block from merged config files. User-authored content is never touched.

What gets installed

Bucket	What it is
Rules (source/rules/)	Always-on guidance: workflow, manual proof before done, code quality, design principles, observability, privacy by design, skills catalog
Skills (source/skills/)	On-demand expertise loaded from the skills catalog: TDD, bug-first, hexagonal, REST, UoW, code review, threat modeling, caching, rate limiting, and more
Commands (source/commands/)	Slash-command prompts: /spec, /plan, /tdd, /bug, /review, /done, /commit, /threat-model, /refactor; /done requires recorded changed-surface manual proof

For the philosophy behind what is installed, see docs/style-overview.md.

Repo layout

style/
├── source/           single source of truth
│   ├── rules/        always-on guidance
│   ├── skills/       on-demand expertise
│   ├── commands/     slash-command prompts
│   └── evals/        static skill-routing assertions (linted, not executed)
├── scripts/          loader, lint, build, per-agent adapters
├── tests/            snapshot tests for adapters
├── docs/             newcomer-facing docs + ADRs
├── AGENTS.md         contributor rules for AI agents working on this repo
└── README.md

Authoring

See docs/contributing.md and AGENTS.md.

Requirements

• Python 3.10+.
• pyyaml (the installer will pip install --user pyyaml if it's missing).
• Optional: Quint on PATH. When present, every PR review, design/spec, test-design task, and TDD cycle creates and executes a fresh task-specific model under ignored ~/.style/specs/quint/. Without it, agents perform and record logical behavioral analysis instead. /specs/quint/ is reserved by this repo's .gitignore when checkout itself lives at ~/.style; scratch under a separate ~/.style is outside this worktree.