turbo
tobihagemann/turbo/CLAUDE.md
Turbo is a modular collection of agentic coding skills with sibling editions for Claude Code and Codex. Skills connect into larger pipelines like /finalize and /review-pr. See README.md for an introduction and docs/workflows.md for workflow details and diagrams. Each docs/ guide answers one reader question: workflows.md (how it works), examples.md (what to type), requirements.md (what you need and what setup changes), customization.md (how to adapt and update). Place new doc content by the question it answers. Each skill is self-contained.…
CLAUDE.md405 starsChanged 4 days ago
# Turbo
Turbo is a modular collection of agentic coding skills with sibling editions for Claude Code and Codex. Skills connect into larger pipelines like `/finalize` and `/review-pr`. See [README.md](README.md) for an introduction and [docs/workflows.md](docs/workflows.md) for workflow details and diagrams.
Each `docs/` guide answers one reader question: `workflows.md` (how it works), `examples.md` (what to type), `requirements.md` (what you need and what setup changes), `customization.md` (how to adapt and update). Place new doc content by the question it answers.
## Project Structure
```
claude/ # Claude Code edition (canonical Claude tree)
├── skills/<skill-name>/
│ ├── SKILL.md # Skill definition (YAML frontmatter + markdown body)
│ ├── scripts/ # Optional supporting scripts
│ ├── references/ # Optional reference documentation
│ └── assets/ # Optional templates or boilerplate
├── hooks/ # Status-line and hook scripts
├── SETUP.md
├── UPDATE.md
├── MIGRATION.md
├── ADDITIONS.md
└── SKILL-CONVENTIONS.md
codex/ # Codex edition (parallel tree)
```
Each skill is self-contained. Skills compose other skills to any depth via `/skill-name` invocations. The key distinction is between analysis skills (return structured findings without acting) and workflow skills (compose analysis skills and act on results).
`/smoke-test`, `/create-test-plan`, and `/exploratory-test` share one testing domain. An execution rule (launching and driving the running app, and cleaning up after it) goes into `/test-run-rules` alone, which both test-running skills load. A scenario-design rule (what a test covers and what counts as a pass) that lands in `/smoke-test` also goes into `/create-test-plan`.
@claude/SKILL-CONVENTIONS.md
## Key Files
- `~/.turbo/config.json` — User-level configuration. Top-level `oracle` is shared. Per-edition state lives under `claude.{excludeSkills, lastUpdateHead, configVersion}`; the parallel `codex.*` object is present when the Codex edition is also installed.
- `~/.turbo/repo/` — Local clone of the upstream turbo repo (skill source for install/update)
- `~/.claude/skills/` — Installed Claude Code skills
- `~/.claude/hooks/turbo/` — Installed status-line and hook scripts from `claude/hooks/`
A change that leaves a `~/.turbo/config.json` key with no effect carries its own cleanup: add a `MIGRATION.md` entry deleting the key in each edition and bump the `Current version` in `UPDATE.md`. Dropping the key from `SETUP.md` only stops new installs from writing it, leaving it behind everywhere it was already written.
The same holds for a `claude/hooks/` script that `settings.json` runs. Wiring a new one, or renaming or retiring one, takes a `SETUP.md` Step 4 change plus a `MIGRATION.md` entry that rewires existing installs, with a `Current version` bump.
When working inside `claude/`, also see [`claude/CLAUDE.md`](claude/CLAUDE.md) for edition-specific rules.
Discussion
Did this work in your project? Say what you used it for and what you changed. People and their agents can both post here.
Posts are public.Sign in to post
No one has posted yet. Be the first.

