Skip to content
Agentic Levels
Start hereLevelsTracksPromptsAbout
Get my levelGet my level

  • Start hereYour first useful task with AI.
  • Level quiz8 questions. No AI needed.
  • Fluency ScoreYour Level and a score, from your own AI.

  • LevelsThe 11 levels, in order.
  • LessonsShort lessons, filter by level.
  • TracksGuided paths by tool, role or goal.
  • Set up your AI tool
  • Starter packFree files that help AI remember.

  • PromptsReady prompts to copy and adapt.
  • ChatGPT vs Claude vs GeminiPick the right one for the job.
  • How-tos
  • ReferenceCommands for Claude Code, Codex, Antigravity.
  • ResourcesHand-picked reading and tools.

  • Your progress
  • Your next moveAn interview that picks your next skill.
  • Your next lessonsThe next lessons at your level.
  • Updates

  • Pricing
  • About
  • FAQ & Feedback
  • Work with us
  • How we handle your data
Sign inPreferences
AboutPricingHow-tosPromptsStarter packMy stackHow we handle your data

© 2026 Fuentes Studio

PrivacyTermsRefundsAccessibilityAI content policyCredits
yourCouncil
Ready to help
✦

What do you want to understand?

Ask anything about what you're learning.

←How-to guides
  1. Resources
  2. ›How-to guides

How to use Claude Code Prompt Improver

Hook-based prompt enrichment for Claude Code, automatically asks 1-6 clarifying questions for vague requests before execution, achieving 31% token reduction.

Levels 2 to 4On GitHub

Vague task briefs cause Claude Code to go off in the wrong direction, requiring correction messages that cost more tokens than a clarifying question upfront would have. Prompt Improver installs as a PostToolUse hook that evaluates every request for clarity. Clear requests proceed immediately. Ambiguous ones trigger 1-6 targeted clarifying questions before Claude touches any code, resulting in 31% token reduction across sessions. Bypass prefixes (*, /, #) let you skip evaluation for commands where you want immediate execution. Requires Claude Code 2.0.22+.

Before you start

Claude Code 2.0.22+ installed. The tool installs as a hook. Verify hook support with claude --version.

  1. 1Install the hook into Claude Code settings
  2. 2Verify the hook fires on an ambiguous request
  3. 3Use bypass prefixes when you want immediate execution
  4. 4Monitor token savings over time

Steps

01

Install the hook into Claude Code settings

Prompt Improver works as a PostToolUse hook. Add it to your Claude Code settings.json hooks section. After installation, restart Claude Code, every subsequent request will pass through the clarity evaluator before execution.

Add to Claude Code hooks
# Clone or install: git clone https://github.com/severity1/claude-code-prompt-improver # Add to ~/.claude/settings.json hooks section: { "hooks": { "PostToolUse": [{ "command": "[PATH_TO_IMPROVER]/evaluate-prompt.sh" }] } }
[PATH_TO_IMPROVER]Path where you cloned the repo e.g. /Users/me/.claude/tools/claude-code-prompt-improver
02

Verify the hook fires on an ambiguous request

Write a deliberately vague task, something like 'clean up the auth module.' The hook should intercept it and ask 1-6 clarifying questions before proceeding. If it passes through without questions, check that the hook path is correct in settings.json.

Test with a vague request
Clean up the auth module.
03

Use bypass prefixes when you want immediate execution

Three prefixes bypass the clarity evaluator: * (star), / (slash), # (hash). Prefix your request with any of these to skip the clarifying questions and execute immediately. Use these for commands where you know exactly what you want and the evaluation overhead is unnecessary.

Bypass with prefix
* [YOUR_REQUEST_THAT_SHOULD_SKIP_EVALUATION]
[YOUR_REQUEST_THAT_SHOULD_SKIP_EVALUATION]Request to execute immediately e.g. run the test suite
04

Monitor token savings over time

The hook logs clarity evaluations and bypass decisions. After a week of use, review the log to see which request types consistently needed clarification: these are your highest-leverage areas for writing better briefs upfront.

How do you know it worked?

  • The hook fires on an ambiguous request and asks at least one clarifying question before executing.
  • Clear requests pass through without triggering clarification.
  • Bypass prefixes (*, /, #) skip evaluation immediately.
  • After a week of use, first-pass execution is more accurate than before.

If something looks off

Requires Claude Code 2.0.22+
The hook uses PostToolUse events introduced in Claude Code 2.0.22. Run `claude --version`. If it shows an older version, update before installing.
Hook path must be absolute
Relative paths in settings.json hooks do not resolve correctly. Use the full absolute path to the evaluate-prompt.sh script.
Bypass prefixes apply to the whole request
A * prefix skips evaluation for the entire request, not just the first sentence. Use it only when the whole request is genuinely clear and you want immediate execution.

See also

  • Anthropic prompt engineering tutorial
  • Claude Code best practices

What you'll do, 4 steps

  1. 1Install the hook into Claude Code settings
  2. 2Verify the hook fires on an ambiguous request
  3. 3Use bypass prefixes when you want immediate execution
  4. 4Monitor token savings over time

Was this helpful?