Every heavy Claude user ⭐ Start here — essential

Optimizer Pack

Ten skills aimed at one thing: making Claude cheaper and faster to run — token budget audits, context compression, prompt-cache optimization, persistent memory, agent-history search. Verified workhorses plus the freshest efficiency finds from our crawler (queue-marked), with our token-optimization checklist.

  • 10 skills focused on cost and context efficiency
  • Memory Management (9.2/10) + Systematic Debugging (9.6/10) as verified anchors
  • Token-optimization checklist — the audit order that finds the money
  • Setup guide with the correct install order
  • Lifetime updates — queue skills get verdicts, pack gets refreshed

What you actually download

In the zip

  • SETUP-GUIDE.md — install order + test verdicts
  • install.sh — one-command setup
  • templates/token-optimization-checklist.md
  • templates/CLAUDE-baseline.md
Preview the SETUP-GUIDE.md (first page)
# Optimizer Pack — Setup Guide

> Every skill below was tested on real work by SkillProof. Scores are out of 10;
> full test reports live at https://skillproof.dev/skills/<slug>.

## Install order (it matters — dependencies first)

### 1. Token Budget Auditor — in test queue
Find what actually burns your tokens and fix the top offenders.

### 2. Context Compression — in test queue
Summarize and restructure context to cut token burn in long sessions.

### 3. Prompt Cache Optimizer — in test queue

### 4. Memory Management — ★ 9.2/10

Every skill inside, with its test verdict

  1. 001

    Cost Tracker

    Static reference for what drives Claude Code session cost and per-task-type budget thresholds — not a transcript auditor.

    Efficiency by rohitg00 tested Jul 14, 2026

    8.0/10 Tested · Works
  2. 002

    Compact Guard

    Pre/post-compaction checklist that preserves task state inside Claude Code's real compaction limits.

    Efficiency by rohitg00 tested Jul 14, 2026

    8.8/10 Tested · Works
  3. 003

    Prompt Caching

    Cross-provider (Anthropic/OpenAI/Bedrock/Gemini) prompt-caching diagnostic playbook — four of its own linked reference files are missing from the repo.

    Efficiency by cobusgreyling tested Jul 14, 2026

    7.2/10 Works with setup
  4. 004

    Token Doctor

    Reads your local Claude Code/Cowork transcripts and prints a real cost-shape diagnosis in the terminal.

    Efficiency by techwolf-ai tested Jul 13, 2026

    9.2/10 Tested · Works
  5. 005

    Compact Memory Implementation

    Developer guide to implement compact memory/context compaction in Claude agents.

    Coding by simbajigege tested Jul 13, 2026

    9.2/10 Tested · Works
  6. 006

    Task Profile

    Mines your Claude Code/Cowork session history into a task profile, coaching panel, and skill proposals.

    Productivity by techwolf-ai tested Jul 13, 2026

    9.6/10 Tested · Works
  7. 007

    Clean Cache

    Tiered, confirmation-gated cache cleanup for Flutter/Android/iOS/Node workspaces — verified not to touch lockfiles or source.

    Efficiency by addxai tested Jul 14, 2026

    9.6/10 Tested · Works
  8. 009

    Systematic Debugging

    Stops guess-and-check fixes; enforces hypothesis-driven debugging.

    Coding by Jesse Vincent tested Jun 25, 2026

    9.6/10 Tested · Works
  9. 010

    SkillOpt-Sleep

    Offline nightly sleep-cycle that mines past sessions and gates CLAUDE.md/SKILL.md edits.

    Efficiency by microsoft tested Jul 10, 2026

    9.2/10 Tested · Works

From the test bench

First-party notes from assembling and re-testing this pack — not customer reviews. Buyer feedback is published here as it comes in (reply to your purchase email to add yours).

  • ★★★★☆

    Honest status: 7 of the 10 skills here are crawler finds still in our test queue — the before/after billing comparison and cache-hit-rate measurements are running now. The anchors are verified: Memory Management 9.2, Systematic Debugging 9.6, Skill Creator 9.6.

    SkillProof test bench · verified during pack assembly

  • ★★★★☆

    Claude-Mem came in via the crawler at 85,685 GitHub stars and is queued, not yet verified. What we can stand behind today is the token-optimization checklist — the audit order that finds the money — and the verified anchors; queue skills get verdicts and the pack gets refreshed.

    SkillProof test bench · verified during pack assembly