claude-code-configuration

About

My personal Claude Code configuration.

Approach

• Manual invocation only. Every skill explicitly opts out of auto-loading. The default Claude Code behavior is to auto-select skills whose descriptions look relevant to the conversation, and I'd rather control which mode I'm in.

• Workflow-driven. Each skill is a workflow for one mode of work: shipping changes, reviewing code, writing tests, writing tickets, discussing product...

• Global hard limits. Universal guardrails applied across all skills, found in CLAUDE.md, which Claude Code auto-loads for every session.

• Bypass permissions by default. settings.json sets permissions.defaultMode: bypassPermissions so Claude Code runs without prompting before each tool call. Work happens in version-controlled repositories, so git diff shows me exactly what changed, I found that approvals were more annoying than safe, so I turn them off.

Skills

• /database — database design, querying, indexing, transactions; relational by default, scope-tagged for non-relational
• /devops — DevOps and infrastructure mentorship
• /draft-logo — drafting a logo mark for a project via an iterative HTML gallery of SVG concepts
• /handoff — handing off a session's context to the next agent. Name inspired from Matt Pockock
• /linux — Linux internals and operations
• /meta — authoring Claude Code configuration (skills, CLAUDE.md, settings, hooks, agents, plugins)
• /product — product and UX decisions
• /review-code — structured code/PR feedback
• /ship — extending and fixing production code
• /test — testing against requirements
• /write-task — turning work into a well-scoped development ticket

Invoke any of them by typing /<name> in Claude Code. They never run unless you ask.

Install

git clone [email protected]:yamanidev/claude-code-configuration.git
cd claude-code-configuration
./install.sh

install.sh symlinks three things into a Claude Code config directory:

• skills/ → <config>/skills
• CLAUDE.md → <config>/CLAUDE.md
• settings.json → <config>/settings.json

With no arguments, it targets the default ~/.claude. Pass one or more paths to target a different config dir(s):

./install.sh ~/.claude-work ~/.claude-personal

That's the form to use if you run multiple Claude Code accounts side-by-side via CLAUDE_CONFIG_DIR, the setup I describe in How to Use Multiple Claude Code Accounts (Work & Personal).

To stay up-to-date with this configuration, just run git pull, updates propagate through the symlinks automatically.

Run ./install.sh --help for usage details.

Status line

The script also offers a statusline customization, requires jq. Off by default, pass --statusline to install.sh to enable it.

• Line 1: Active model and version, with the configuration directory name appended for multi-account setups
• Line 2: Usage for both the 5 hour and 7 days windows (color-coded by threshold), and current-session context as an absolute token count (green < 80k, yellow 80–160k, red ≥ 160k) as a heuristic tracking for quality degradation, independent of whether the model advertises a 200k or 1M window

License

MIT