Now liveThe Skillselion MCP - thousands of ranked skills, loaded into your agent mid-task. No install.Get it →
alexgreensh avatar

Token Optimizer

alexgreensh/token-optimizer

  • 514 installs
  • 1.8k repo stars
  • Updated August 3, 2026
  • alexgreensh/token-optimizer

Token Optimizer is a Claude Code skill that audits an agent setup, finds context-window token waste, applies config fixes, and measures the recovered context.

About

Token Optimizer audits a Claude Code or Codex setup, identifies where context-window tokens are wasted, implements config fixes, and measures the resulting savings. A developer runs it when a session's context feels tight and they want to recover room. It supports multiple runtimes (Claude Code, Codex, OpenCode, Copilot) via a runtime gate that picks the correct workflow.

  • Audits Claude Code/Codex config to find context-window waste
  • Implements fixes and measures the token savings
  • Targets 5-15% context recovery, up to 25%+ with autocompact management

Token Optimizer by the numbers

  • 514 all-time installs (skills.sh)
  • Ranked #754 of 3,282 Productivity & Planning skills by installs in the Skillselion catalog
  • Data as of Aug 5, 2026 (Skillselion catalog sync)
At a glance

token-optimizer capabilities & compatibility

Capabilities
token optimization · config audit
Use cases
token optimization
Platforms
macOS
Pricing
Free
npx skills add https://github.com/alexgreensh/token-optimizer --skill token-optimizer

Add your badge

Show developers this skill is listed on Skillselion. Paste this into your README.

Listed on Skillselion
Installs514
repo stars★ 1.8k
Last updatedAugust 3, 2026
Repositoryalexgreensh/token-optimizer ↗

What it does

Audit a Claude Code or Codex setup to find and remove context-window waste, then measure the token savings.

Who is it for?

Claude Code or Codex users who want to recover context-window space and cut token cost

Skip if: one-off scripting tasks or runtimes it does not support

When should I use this skill?

when context feels tight or you want to see where an agent's context window is going

What you get

A cleaned-up config that recovers 5-15% of the context window, up to 25%+ with autocompact management, with measured before/after savings.

  • context-waste audit findings
  • config fixes
  • measured before/after token savings

By the numbers

  • Target 5-15% context recovery through config cleanup
  • up to 25%+ with autocompact management
  • 5-phase audit workflow (Quick Audit, Analysis, Present Findings, Implementation, Verification)

Files

assets/active-compression-hero.svgTextGitHub ↗
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1000 800" width="1000" height="800" role="img" aria-labelledby="title desc">
  <title id="title">Active Compression — nine features, all automatic</title>
  <desc id="desc">Nine compression features, all on by default and handled automatically by hooks: delta mode, structure map, bash compression, search compression, lean-output nudges, quality nudges, loop detection, activity mode, decision extraction.</desc>

  <defs>
    <linearGradient id="bg" x1="0" y1="0" x2="1" y2="1">
      <stop offset="0%" stop-color="#101318"/>
      <stop offset="58%" stop-color="#171b24"/>
      <stop offset="100%" stop-color="#0b0e13"/>
    </linearGradient>
    <linearGradient id="headGlow" x1="0" y1="0" x2="1" y2="0">
      <stop offset="0%" stop-color="#7dd3fc"/>
      <stop offset="48%" stop-color="#a7f3d0"/>
      <stop offset="100%" stop-color="#c4b5fd"/>
    </linearGradient>
    <linearGradient id="gGreen" x1="0" y1="0" x2="1" y2="0"><stop offset="0%" stop-color="#16a34a"/><stop offset="100%" stop-color="#86efac"/></linearGradient>
    <linearGradient id="gEmerald" x1="0" y1="0" x2="1" y2="0"><stop offset="0%" stop-color="#059669"/><stop offset="100%" stop-color="#6ee7b7"/></linearGradient>
    <linearGradient id="gCyan" x1="0" y1="0" x2="1" y2="0"><stop offset="0%" stop-color="#0891b2"/><stop offset="100%" stop-color="#67e8f9"/></linearGradient>
    <linearGradient id="gSky" x1="0" y1="0" x2="1" y2="0"><stop offset="0%" stop-color="#2563eb"/><stop offset="100%" stop-color="#93c5fd"/></linearGradient>
    <linearGradient id="gAmber" x1="0" y1="0" x2="1" y2="0"><stop offset="0%" stop-color="#d97706"/><stop offset="100%" stop-color="#fcd34d"/></linearGradient>
    <linearGradient id="gRose" x1="0" y1="0" x2="1" y2="0"><stop offset="0%" stop-color="#e11d48"/><stop offset="100%" stop-color="#fda4af"/></linearGradient>
    <linearGradient id="gPurple" x1="0" y1="0" x2="1" y2="0"><stop offset="0%" stop-color="#7c3aed"/><stop offset="100%" stop-color="#c4b5fd"/></linearGradient>
    <linearGradient id="gIndigo" x1="0" y1="0" x2="1" y2="0"><stop offset="0%" stop-color="#4f46e5"/><stop offset="100%" stop-color="#a5b4fc"/></linearGradient>
    <linearGradient id="gViolet" x1="0" y1="0" x2="1" y2="0"><stop offset="0%" stop-color="#9333ea"/><stop offset="100%" stop-color="#d8b4fe"/></linearGradient>
    <filter id="softShadow" x="-12%" y="-12%" width="124%" height="140%">
      <feDropShadow dx="0" dy="12" stdDeviation="16" flood-color="#000000" flood-opacity="0.34"/>
    </filter>

    <style>
      .ui { font-family: Inter, ui-sans-serif, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif; }
      .muted { fill: #cbd5e1; }
      .quiet { fill: #94a3b8; }
      .strong { fill: #f8fafc; }
      .head { animation: fadeHead 22000ms ease-in-out infinite; }
      .card { animation-duration: 22000ms; animation-timing-function: cubic-bezier(.2,.9,.2,1); animation-iteration-count: infinite; opacity: 1; }
      .c1 { animation-name: rise1; } .c2 { animation-name: rise2; } .c3 { animation-name: rise3; }
      .c4 { animation-name: rise4; } .c5 { animation-name: rise5; } .c6 { animation-name: rise6; }
      .c7 { animation-name: rise7; } .c8 { animation-name: rise8; } .c9 { animation-name: rise9; }
      @keyframes fadeHead { 0%, 97%, 100% { opacity: 0; } 5%, 92% { opacity: 1; } }
      @keyframes rise1 { 0%, 6%, 97%, 100% { opacity: 0; transform: translateY(16px); } 13%, 92% { opacity: 1; transform: translateY(0); } }
      @keyframes rise2 { 0%, 9%, 97%, 100% { opacity: 0; transform: translateY(16px); } 16%, 92% { opacity: 1; transform: translateY(0); } }
      @keyframes rise3 { 0%, 12%, 97%, 100% { opacity: 0; transform: translateY(16px); } 19%, 92% { opacity: 1; transform: translateY(0); } }
      @keyframes rise4 { 0%, 15%, 97%, 100% { opacity: 0; transform: translateY(16px); } 22%, 92% { opacity: 1; transform: translateY(0); } }
      @keyframes rise5 { 0%, 18%, 97%, 100% { opacity: 0; transform: translateY(16px); } 25%, 92% { opacity: 1; transform: translateY(0); } }
      @keyframes rise6 { 0%, 21%, 97%, 100% { opacity: 0; transform: translateY(16px); } 28%, 92% { opacity: 1; transform: translateY(0); } }
      @keyframes rise7 { 0%, 24%, 97%, 100% { opacity: 0; transform: translateY(16px); } 31%, 92% { opacity: 1; transform: translateY(0); } }
      @keyframes rise8 { 0%, 27%, 97%, 100% { opacity: 0; transform: translateY(16px); } 34%, 92% { opacity: 1; transform: translateY(0); } }
      @keyframes rise9 { 0%, 30%, 97%, 100% { opacity: 0; transform: translateY(16px); } 37%, 92% { opacity: 1; transform: translateY(0); } }
      @media (prefers-reduced-motion: reduce) {
        .head, .card { animation: none; opacity: 1; transform: none; }
      }
      .card { transform-box: fill-box; }
    </style>
  </defs>

  <rect width="1000" height="800" rx="28" fill="url(#bg)"/>
  <path d="M70 120 C240 56 380 76 500 150 C636 230 770 214 936 116" fill="none" stroke="#334155" stroke-width="2" opacity=".30"/>

  <!-- Header -->
  <g class="head ui">
    <text x="500" y="74" class="strong" font-size="44" font-weight="850" text-anchor="middle">Active Compression</text>
    <text x="500" y="112" class="muted" font-size="21" font-weight="560" text-anchor="middle">Nine features · all on by default · the hooks handle everything</text>
  </g>

  <!-- ============ ROW 1 ============ -->
  <!-- Card 1: Delta Mode -->
  <g class="card c1 ui" filter="url(#softShadow)">
    <rect x="36" y="152" width="294" height="186" rx="18" fill="#1c2230" stroke="#16a34a" stroke-width="1.6" opacity="0.9"/>
    <text x="60" y="194" fill="#86efac" font-size="22" font-weight="850">Delta Mode</text>
    <text x="60" y="232" class="muted" font-size="16" font-weight="520">Re-reads return only the diff.</text>
    <text x="60" y="256" class="muted" font-size="16" font-weight="520">65%+ of reads are re-reads.</text>
    <rect x="60" y="280" width="246" height="40" rx="11" fill="#0f2417"/>
    <text x="183" y="306" fill="#bbf7d0" font-size="17" font-weight="850" text-anchor="middle">~20% on re-reads</text>
  </g>

  <!-- Card 2: Structure Map -->
  <g class="card c2 ui" filter="url(#softShadow)">
    <rect x="353" y="152" width="294" height="186" rx="18" fill="#1c2230" stroke="#059669" stroke-width="1.6" opacity="0.9"/>
    <text x="377" y="194" fill="#6ee7b7" font-size="22" font-weight="850">Structure Map</text>
    <text x="377" y="232" class="muted" font-size="16" font-weight="520">Code re-reads return a skeleton:</text>
    <text x="377" y="256" class="muted" font-size="16" font-weight="520">signatures, imports, classes.</text>
    <rect x="377" y="280" width="246" height="40" rx="11" fill="#082f24"/>
    <text x="500" y="306" fill="#a7f3d0" font-size="17" font-weight="850" text-anchor="middle">up to 99% per file</text>
  </g>

  <!-- Card 3: Bash Compression -->
  <g class="card c3 ui" filter="url(#softShadow)">
    <rect x="670" y="152" width="294" height="186" rx="18" fill="#1c2230" stroke="#0891b2" stroke-width="1.6" opacity="0.9"/>
    <text x="694" y="194" fill="#67e8f9" font-size="22" font-weight="850">Bash Compression</text>
    <text x="694" y="232" class="muted" font-size="16" font-weight="520">CLI output to compact summary.</text>
    <text x="694" y="256" class="muted" font-size="16" font-weight="520">Credentials always preserved.</text>
    <rect x="694" y="280" width="246" height="40" rx="11" fill="#082832"/>
    <text x="817" y="306" fill="#a5f3fc" font-size="17" font-weight="850" text-anchor="middle">~10% on CLI output</text>
  </g>

  <!-- ============ ROW 2 ============ -->
  <!-- Card 4: Search Compression -->
  <g class="card c4 ui" filter="url(#softShadow)">
    <rect x="36" y="358" width="294" height="186" rx="18" fill="#1c2230" stroke="#2563eb" stroke-width="1.6" opacity="0.9"/>
    <text x="60" y="400" fill="#93c5fd" font-size="22" font-weight="850">Search Compression</text>
    <text x="60" y="438" class="muted" font-size="16" font-weight="520">grep/web results to top hits</text>
    <text x="60" y="462" class="muted" font-size="16" font-weight="520">plus counts, not the full dump.</text>
    <rect x="60" y="486" width="246" height="40" rx="11" fill="#0c1b3a"/>
    <text x="183" y="512" fill="#bfdbfe" font-size="17" font-weight="850" text-anchor="middle">~15% on search</text>
  </g>

  <!-- Card 5: Lean-Output Nudges -->
  <g class="card c5 ui" filter="url(#softShadow)">
    <rect x="353" y="358" width="294" height="186" rx="18" fill="#1c2230" stroke="#d97706" stroke-width="1.6" opacity="0.9"/>
    <text x="377" y="400" fill="#fcd34d" font-size="22" font-weight="850">Lean-Output Nudges</text>
    <text x="377" y="438" class="muted" font-size="16" font-weight="520">Steers to concise output when</text>
    <text x="377" y="462" class="muted" font-size="16" font-weight="520">context fills past 55%.</text>
    <rect x="377" y="486" width="246" height="40" rx="11" fill="#3a2408"/>
    <text x="500" y="512" fill="#fde68a" font-size="17" font-weight="850" text-anchor="middle">30–41% output cut</text>
  </g>

  <!-- Card 6: Quality Nudges -->
  <g class="card c6 ui" filter="url(#softShadow)">
    <rect x="670" y="358" width="294" height="186" rx="18" fill="#1c2230" stroke="#e11d48" stroke-width="1.6" opacity="0.9"/>
    <text x="694" y="400" fill="#fda4af" font-size="22" font-weight="850">Quality Nudges</text>
    <text x="694" y="438" class="muted" font-size="16" font-weight="520">Warns when context quality</text>
    <text x="694" y="462" class="muted" font-size="16" font-weight="520">drops 15+ points.</text>
    <rect x="694" y="486" width="246" height="40" rx="11" fill="#3a0c18"/>
    <text x="817" y="512" fill="#fecdd3" font-size="17" font-weight="850" text-anchor="middle">prevents quality loss</text>
  </g>

  <!-- ============ ROW 3 ============ -->
  <!-- Card 7: Loop Detection -->
  <g class="card c7 ui" filter="url(#softShadow)">
    <rect x="36" y="564" width="294" height="186" rx="18" fill="#1c2230" stroke="#7c3aed" stroke-width="1.6" opacity="0.9"/>
    <text x="60" y="606" fill="#c4b5fd" font-size="22" font-weight="850">Loop Detection</text>
    <text x="60" y="644" class="muted" font-size="16" font-weight="520">Catches retry loops before</text>
    <text x="60" y="668" class="muted" font-size="16" font-weight="520">they burn tokens.</text>
    <rect x="60" y="692" width="246" height="40" rx="11" fill="#23123a"/>
    <text x="183" y="718" fill="#ddd6fe" font-size="17" font-weight="850" text-anchor="middle">measured per loop</text>
  </g>

  <!-- Card 8: Activity Mode -->
  <g class="card c8 ui" filter="url(#softShadow)">
    <rect x="353" y="564" width="294" height="186" rx="18" fill="#1c2230" stroke="#4f46e5" stroke-width="1.6" opacity="0.9"/>
    <text x="377" y="606" fill="#a5b4fc" font-size="22" font-weight="850">Activity Mode</text>
    <text x="377" y="644" class="muted" font-size="16" font-weight="520">Classifies session live:</text>
    <text x="377" y="668" class="muted" font-size="16" font-weight="520">code / debug / review / infra.</text>
    <rect x="377" y="692" width="246" height="40" rx="11" fill="#181a3a"/>
    <text x="500" y="718" fill="#c7d2fe" font-size="17" font-weight="850" text-anchor="middle">adapts compaction</text>
  </g>

  <!-- Card 9: Decision Extraction -->
  <g class="card c9 ui" filter="url(#softShadow)">
    <rect x="670" y="564" width="294" height="186" rx="18" fill="#1c2230" stroke="#9333ea" stroke-width="1.6" opacity="0.9"/>
    <text x="694" y="606" fill="#d8b4fe" font-size="22" font-weight="850">Decision Extraction</text>
    <text x="694" y="644" class="muted" font-size="16" font-weight="520">Captures decisions in real</text>
    <text x="694" y="668" class="muted" font-size="16" font-weight="520">time, kept through compaction.</text>
    <rect x="694" y="692" width="246" height="40" rx="11" fill="#2a0f3a"/>
    <text x="817" y="718" fill="#e9d5ff" font-size="17" font-weight="850" text-anchor="middle">survives compaction</text>
  </g>
</svg>

Related skills

mattpocockGrill MeHave an agent relentlessly pressure-test their plan or design until every decision branch is resolved and shared understanding is reached.691k196kmattpocockGrill With DocsPressure-test architecture decisions against their live codebase and living documentation before writing production code.585k196kopen.feishu.cnLark CalendarLet their coding agents read, create, and manage events inside Lark (Feishu) Calendar directly from natural language instructions.493kopen.feishu.cnLark ImLet their coding agent send instant notifications, post updates, and receive commands directly inside Lark (Feishu) workspaces.493kopen.feishu.cnLark ContactLet their AI coding agents read and write contacts, send messages, and manage user data inside Lark (Feishu).493kopen.feishu.cnLark AttendanceAutomatically fetch and report daily attendance data from Lark (Feishu) into their own tools or agents.493k

How it compares

Pick token-optimizer for full-session config, compaction, and skill-bloat audits; use output-only compressors when you only need to shrink individual command results.

FAQ

What does Token Optimizer do?

It audits a Claude Code or Codex setup, finds context-window waste, implements config fixes, and measures the token savings.

How much context can it recover?

It targets 5-15% context recovery through config cleanup, and up to 25%+ with autocompact management.

Which runtimes does it support?

It gates on runtime and supports Claude Code, Codex, OpenCode, and GitHub Copilot, following the correct workflow for each.

Productivity & Planningworkflow
Browse by category
AI & AgentsFrontend & DesignBackend & DataDev ToolsAutomationTesting & ReviewSecurityMonitoring & CloudDocs & Planning
Design & frontend
Design systemsAnimation & motionAccessibility (a11y)CSS & TailwindUI designUXDesign to codeshadcn/uiLanding pagesBest design skills
Most searched skills
find-skillscavemanfrontend-designbrainstormingsystematic-debuggingstop-slopemil-design-engui-ux-pro-maxkarpathy-guidelinesweb-design-guidelinesreact-best-practicesgrill-meimprovemicrosoft-foundryagent-browser
Browse all listings
All agent skillsAll MCP serversAll marketplacesAll plugins

This week in AI coding

Five minutes, every Monday - the tools, releases and tactics for developers.

unsubscribe anytime.