[ Claude Code · Efficiency Roadmap ]

Mastering Claude Code: The 90% Efficiency Roadmap

A 12-part field guide to slashing token spend in Claude Code — from context hygiene and surgical file referencing to prompt caching, model routing, and measuring real cost per feature.

12 lessons1 published~84 min total
  1. 01
    FoundationsPublished

    Context Hygiene

    Stale context is dead weight you re-pay for every turn. Master /clear and /compact to reset the running bill.

    Read lesson →
  2. 02
    FoundationsComing Soon

    The Token Budget Mental Model

    The context window is RAM, and every message re-sends the history. See the meter before you spend.

  3. 03
    PrecisionComing Soon

    Surgical File Referencing

    Don't dump the repo — point at lines. @-mentions over paste, so Claude reads one file, not forty.

  4. 04
    PrecisionComing Soon

    CLAUDE.md as a Context Cache

    Stop re-explaining the project every session. Amortize your setup tokens to near zero.

  5. 05
    PrecisionComing Soon

    Custom Slash Commands & Skills

    Turn a repeated paragraph of instructions into one cheap call. Kill prompt boilerplate for good.

  6. 06
    WorkflowComing Soon

    Plan Mode & Spec-First

    Think before tokens. A wrong turn is the most expensive output there is — plan to avoid re-generation.

  7. 07
    WorkflowComing Soon

    Subagents & Delegation

    Offload search and exploration to a child context so the main window stays lean and cheap.

  8. 08
    WorkflowComing Soon

    Model Routing

    Don't pay Opus rates for a rename. Match Haiku, Sonnet, and Opus to the task in front of you.

  9. 09
    ScaleComing Soon

    Prompt Caching

    Structure prompts so the static prefix caches — and take roughly 90% off those tokens on every reuse.

  10. 10
    ScaleComing Soon

    MCP Servers over Data Dumps

    Give Claude a tool, not a paste. Fetch data on demand instead of front-loading the context window.

  11. 11
    ScaleComing Soon

    Hooks & Automation

    Deterministic guardrails kill correction round-trips — the cheapest token is the one you never send.

  12. 12
    ScaleComing Soon

    Measuring ROI

    /cost, ccusage, and a token budget per feature. What you don't measure, you overpay for.

Burning tokens faster than budget?

This is the workflow I use to keep Claude Code lean. Let's tune yours.

Start a conversation →