agentsclimarketplace

Skill budget

Skill dhruba-datta/claude-papercuts/skills/skill-budget

Ten Claude Code skills that fix bugs Anthropic closed as 'not planned'.

Install
npx -y skills add dhruba-datta/claude-papercuts --skill skill-budget

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

One thing to look at

  • 0 stars0 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.

What its author says it does

Copied from the file, not written here

Audit installed Claude Code skills against the system-prompt char budget and show which ones may have silently fallen out of the model's visible list. Use this skill when the user runs /claude-papercuts:skill-budget, asks why their skills stopped triggering, mentions installed-but-invisible skills, or wants a Monday-morning health check of their Claude setup. Runs the audit.py script in this skill's directory and presents the result.

SKILL.md

4.0 KB, 868 tokens by cl100k_base, as published. Nobody here has run it

skill-budget — audit your installed skills

Fixes: #30387, #34648, #16575

What this prevents

"In environments with many skills (40+) and rich configuration, the model bypasses the Skill tool entirely and uses built-in tools directly, even when skill descriptions explicitly state 'ALWAYS use this skill.'" — issue #34648

You install 40 skills. The system prompt has a limited budget for skill descriptions. The excess falls off silently — no warning, no error. Claude can't see them, and you don't know it.

What this skill does

When invoked, run the bundled audit.py script (next to this SKILL.md) and present its output to the user verbatim.

The script scans three locations:

  • ~/.claude/skills/<name>/SKILL.md — your global skills
  • .claude/skills/<name>/SKILL.md — project-level skills
  • ~/.claude/plugins/**/skills/<name>/SKILL.md — plugin-installed skills

It parses each SKILL.md's YAML frontmatter, sums the name + description chars (the approximate per-skill cost in the system prompt), and ranks skills by weight. Skills whose cumulative sum exceeds the budget are marked INVISIBLE TO CLAUDE.

How to invoke (the actual procedure)

  1. Run the audit script with the user's current settings:

    ${CLAUDE_PLUGIN_ROOT}/skills/skill-budget/audit.py

  2. If the user passed arguments (e.g. --budget 12000, --json), forward them.

  3. Show the script's output verbatim. Do not summarize or rewrite — the bar chart and the suggested mv ... .disabled commands are the value. Reformatting them defeats the purpose.

  4. After showing the output, if any skills are marked INVISIBLE:

    • Ask the user if they want help deciding which ones to disable.
    • If asked, recommend disabling the largest-char skills first (highest information density per char freed) or the ones the user uses least often.

  5. Never auto-disable a skill without explicit user consent. The suggested mv commands in the output are for the user to inspect and run themselves.

When to auto-invoke

  • User runs /claude-papercuts:skill-budget
  • User asks "are my skills loaded?"
  • User mentions a specific skill not triggering when they expect
  • User says they recently installed N new skills and notices degraded behavior
  • User asks for a "Monday morning check" or "skill audit"

What this skill does NOT do

  • It does not run the actual model to test which skills trigger; it estimates based on char budget. The 15,000-char default is inferred from issue discussions, not officially documented.
  • It does not modify any files. The suggested mv ... .disabled commands are for the user to copy-paste if they choose.
  • It does not include skills that have already been .disabled'd (renamed). Those are off-budget by definition.

Configuration

# Use a different budget (e.g. if you know your environment's cap)
${CLAUDE_PLUGIN_ROOT}/skills/skill-budget/audit.py --budget 12000

# Machine-readable output (for piping into other tools)
${CLAUDE_PLUGIN_ROOT}/skills/skill-budget/audit.py --json

Make it a habit

The retention play: run this every Monday morning, after every skill install, or whenever Claude stops triggering a skill you expect. Same flywheel as npm outdated or brew doctor — a weekly health check.

Gives 0 of the 12 instructions most prompt engineering skills give in 868 tokens

Counted across 563 of the 626 authors here whose files we hold, read 2026-08-06

  • ask at most three clarifying questionsin 22 of 563, across 15 files
  • respond in the user input languagein 14 of 563, across 9 files
  • preserve the original intentin 13 of 563, across 11 files
  • Establish baseline metrics and collect representative examplesin 12 of 563, across 2 files
  • Identify failure modes and prioritize high-impact fixesin 12 of 563, across 2 files
  • Apply prompt and workflow improvements with measurable goalsin 12 of 563, across 2 files
  • Roll back quickly if quality or safety metrics regressin 12 of 563, across 2 files
  • validate changes with tests and roll out in controlled stagesin 12 of 563, across 2 files
  • generate quantitative baseline performance reportsin 12 of 563, across 2 files
  • create representative test scenariosin 12 of 563, across 2 files
  • treat prompts as codein 12 of 563, across 5 files
  • test prompts on diverse inputsin 12 of 563, across 8 files

Said here and by no other author read

  • forward any user arguments to the script
  • show script output verbatim
  • ask the user about disabling invisible skills
  • recommend disabling largest-char skills first

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 328,083. 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.