A ready-to-use kit that cuts the tokens Claude Code and Codex waste on stale context, heavy file reads, noisy test logs and an expensive model doing simple work. Set it up in about 10 minutes.

What’s inside:

• START-HERE: the 10-minute setup, in order of impact
• Lean CLAUDE.md and AGENTS.md templates (they load into every session, so size matters)
• Claude Code settings.json that blocks reads of dependencies, build output, lockfiles and .env files
• Two hooks: stop whole-file reads of huge files, and trim test output to failures only
• Two cheap Haiku subagents: a read-only “scout” and a “test-runner” that return short summaries
• A codebase-map skill and prompt, so agents stop re-exploring your repo every session
• Codex config.toml defaulting to gpt-6-luna, plus a “deep” profile for hard tasks
• Prompts for plan-then-build and fresh-session handoffs
• An optional GLM Coding Plan routing example
• A guide explaining why each step works, with sources and trade-offs

Checked against the Claude Code and Codex docs in October 2026. Instant download (ZIP). Free tools it points to are linked and credited, not bundled.

We don’t promise a specific saving: it depends on how you work. The guide shows you how to measure before and after with /usage, /context and /status.

Refunds: non-refundable after download unless the files are broken.