Leo's Claude Starter workflow
A shared set of skills, agents and hooks for Claude Code: design, phased sprints and notes kept outside the repository.
- 64 ★ on the repository
- Last commit
- BSD-2-Clause
Broad access
Why this level
- The `implement-sprint` skill executes a plan phase by phase with edits and commits, and in the author's example the agent launches without confirmation prompts.
- Hooks run scripts on the machine and constrain the agent's actions.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/leo-s-claude-starterDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The `implement-sprint` skill executes a plan phase by phase with edits and commits, and in the author's example the agent launches without confirmation prompts.
- Hooks run scripts on the machine and constrain the agent's actions.
This is third-party code. Review the repository files before installing.
What it does
Leo's Claude Starter keeps skills and agents that connect to several repositories by symbolic links, so a fix reaches every project at once. There are two tiers. The infrastructure tier gives `note`, `understand`, `consult`, `session` and `remote-session` for any repository. The full tier adds architecture and sprint skills, shared agents, a worker protocol and hooks. The working cycle is a chain of commands: `/understand` loads context, `/arch-design` and `/arch-review` prepare and check a change design, `/create-sprint` turns it into a spec, `/implement-sprint` runs phases in a separate worktree, `/review-sprint` audits the result and `/fold-pending` folds the design into the main documentation. Plans, findings and decisions live outside the repository in a note vault so that code search does not drag in unfinished ideas.
Who it is for. Developers with several related repositories who run design and sprint work in Claude Code and want a ready process to start from.
Good fit when
- Several repositories must share one set of skills and agents without copying.
- A change design is reviewed several times in fresh sessions until an implementer can build it without guessing.
- A sprint runs phase by phase in a separate worktree with a gate and a commit at each step.
- A question about another repository must be asked without loading its code into your session.
- A Claude session must be driven from a phone through remote-control and tmux.
Not a fit when
- The project is one small repository: the full sprint process will outweigh the work.
- Symbolic links are not allowed: the set assumes a shared parent directory for the clones.
- You do not want notes kept outside the repository: architecture skills call `note/cli.py` directly.
Example request
/get-startedLimitations
The author calls the set a personal process rather than a recommendation. Links are relative and require the starter and the project to share a parent directory. The architecture skills and `whats-next` call `note/cli.py` directly, so a replacement store must support the `new`, `list`, `status`, `set` and `path` commands. The adoption contract is described in `ADOPTION.md` and checked by `load.py --check`. In its example the author launches Claude with the `--dangerously-skip-permissions` flag, deliberately turning confirmations off.
How to disable. Delete the symbolic links to skills and agents from the project's `.claude/` and from `~/.claude/skills/`. Remove the hooks from the repository settings and leave the note vault as it is.
FAQ
How do I get started with the starter?
Clone the repository next to your project, create a relative link to the `get-started` skill in the project's `.claude/skills/`, restart Claude Code and run `/get-started`. It asks which tier you want and sets up links, bindings and `CLAUDE.md`.
How do the tiers differ?
The infrastructure tier gives `note`, `understand`, `consult`, `session` and `remote-session` and suits any repository. The full tier adds architecture and sprint skills, agents, a worker protocol and hooks for repositories under active development.
How does the note vault work?
The `note` skill keeps seven note types with their own statuses: findings, features, plans, decisions, research, questions and retros. One vault serves many repositories, and status is a field so links do not break.
What do the hooks do?
They enforce the author's habits: outline a markdown file before reading it, use LSP before grep, let only the orchestrator commit and ask clarifying questions in chat rather than through the choice widget.
How do I ask about another repository?
`/consult <repo> <question>` launches a read-only advisor that knows that repository and relays its answer. The mapping of names to paths lives in `~/.config/consult.json`.
Discussion
No questions or reviews yet.
Related
WorkflowsA starter kit for Claude Code: a CLAUDE.md memory-bank template with ready hooks, subagents and skills you copy into your project
A Claude Code working configuration: a planner with per-milestone review, sub-agents and skills for analysis, refactoring and docs sync
A Claude Code environment installer: CLAUDE.md, rules, skills, agents, hooks and two-layer project memory in one script
Claude Code Blueprint: 12 agents, 19 skills, 16 hooks and 6 CLAUDE.md rules, copy what you need or install a preset
One-command Claude Code environment installer: a wizard sets up agents, rules, commands, skills and hooks in three profiles
Claude Code scaffold with 27 commands, 10 hooks and 25 skills installed globally to set common rules for every project.
README in short
The repository began as a template to copy, and now projects adopt it through links. The ADOPTION.md document serves as the contract between the starter and a project, and the author calls the result a starting point for newcomers.