A Claude Code skill that minimises token and tool usage across any task.
Applies by default. No manual invocation needed.
- Plan before executing — identifies the minimum operations needed before making any tool calls
- Aggregators over crawling — uses single-source tools (npmjs, pypi, crates.io, caniuse, gh CLI) instead of hitting multiple sites
- Parallel by default — batches independent tool calls in one message instead of running them sequentially
- Subagents for research — delegates open-ended research to subagents to keep main context clean
- Read minimally — greps before reading, uses offset/limit for large files, never re-reads what's already in context
- Skip noise directories — ignores node_modules, .git, dist, build, pycache, .venv by default
- Context window awareness — treats context like RAM, front-loads decisions, avoids paging in data you won't use
- Output economy — matches response length to task complexity, no padding
claude plugin install github:C-Moir/claude-lean-ops
Or add to your ~/.claude/settings.json:
{
"extraKnownMarketplaces": {
"lean-ops": {
"source": {
"source": "github",
"repo": "C-Moir/claude-lean-ops"
}
}
},
"enabledPlugins": {
"lean-ops@lean-ops": true
}
}MIT