agentsclimarketplace

Codeburn

Skill clarkhager/cowork-skills/skills/codeburn

Track, analyze, and optimize Claude token usage and costs using CodeBurn. Use this skill whenever Clark asks about token usage, AI spending, session costs, what things have cost, how much he's spent, optimization recommendations, or wants to run any CodeBurn command. Also trigger when Clark says things like "how are my limits", "am I burning through tokens", "what's eating my context", "optimize my setup", "check my usage", "codeburn", or "how much have I used this week/month". This skill installs CodeBurn if it's not already present, then runs the appropriate command and interprets the results.From its SKILL.md

Install
npx -y skills add clarkhager/cowork-skills --skill codeburn

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.0 KB, as published. Nobody here has run it

CodeBurn: Token Usage Tracking and Optimization

CodeBurn reads Claude Code and Claude Desktop session data directly from disk — no proxy, no wrapper — and gives Clark a clear picture of where his tokens and budget are going. It also scans his setup for waste and provides copy-paste fixes.

Step 1: Ensure CodeBurn is installed

Before running any command, check if codeburn is available:

which codeburn

If not found, install it:

npm install -g codeburn

Confirm installation succeeded before proceeding.

Step 2: Determine what Clark wants

Based on the conversation, pick the right command:

Clark's intentCommand to run
General usage overview / "what have I spent"codeburn report --provider claude
Today's usagecodeburn today --provider claude
This monthcodeburn month --provider claude
Last 7 dayscodeburn report -p 7days --provider claude
Find waste / optimizecodeburn optimize --provider claude
Compare modelscodeburn compare --provider claude
Quick one-liner statuscodeburn status --provider claude
Yield (productive vs abandoned sessions)codeburn yield --provider claude
Set plan trackingcodeburn plan set claude-max

If unclear, default to codeburn report --provider claude for a full 7-day overview.

Step 3: Run the command and capture output

Run the command via bash and capture stdout. Example:

codeburn report --provider claude --format json

Use --format json for programmatic parsing when you need to pull specific numbers. Use the plain output for display when the user just wants to see results.

Step 4: Interpret and present findings

Don't just dump raw output. Surface what matters:

  • Total spend this period and how it compares to Clark's Max plan ($200/month)
  • Top cost drivers by project, model, or task type
  • One-shot rate — what percentage of edits the model got right first try
  • Cache hit rate — whether prompt caching is working well
  • Biggest single sessions — which sessions burned the most

For the optimize command specifically:

  • List each finding with its estimated dollar/token savings
  • Pull out the copy-paste fix for each one and show it clearly
  • Prioritize by urgency (CodeBurn ranks them A-F)
  • Flag bloated CLAUDE.md or MEMORY.md files immediately — these cost tokens on every session start and the fix is usually a one-line compress

Step 5: Give a recommendation

After showing the data, add one concrete next step. Examples:

  • "Your biggest waste is [X]. Run this to fix it: [command]"
  • "You've used [Y]% of your Max plan equivalent this month — on track / ahead of pace"
  • "The optimize scan found [N] issues. The highest-impact one is [X] — here's the fix"

Notes

  • CodeBurn tracks both Claude Code (~/.claude/projects/) and Claude Desktop (~/Library/Application Support/Claude/local-agent-mode-sessions/) — Clark uses Cowork which writes to the Desktop path, so both will be captured
  • The plan set claude-max command sets a $200/month baseline for the progress bar — worth doing once if not already configured
  • If Clark wants ongoing visibility, suggest running codeburn status --provider claude at the start of sessions
  • Token counts for Cowork sessions may show as "Claude Desktop" in the provider breakdown

Keep looking

Skills are one crate of 608,475. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.