agentleFS
davekilleendavekilleen/Dex/.agents/skills/prompt-improver/SKILL.md

prompt-improver

Skill509 starsChanged 3 months agoMITView on GitHub

Rewrite a vague prompt into a rich, structured one, with automatic fallback. Use when the user says 'improve this prompt', 'make this prompt better', or hands over a thin instruction. Not for creating a reusable skill; use `create-skill`.

  • Reads credentials

12 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

12 rules. 0 are common, 7 are distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 10 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Check if $PROMPT starts with a flag:

    line 31
  2. Very specific:

    Check if $PROMPT starts with a flag (-p, -v) and extract:

    line 47
  3. Very specific:

    Script available?: Check if .scripts/improve-prompt.cjs exists

    line 56
  4. Distinctive:

    If yes → Use script (calls Anthropic API directly)

    line 57
  5. Distinctive:

    If no → Check for API key

    line 58
  6. Very specific:

    API key available?: Check if ANTHROPIC_API_KEY is set in environment

    line 60
  7. Distinctive:

    If yes → Use Anthropic Messages API inline

    line 61
  8. Distinctive:

    If no → Fall back to current LLM

    line 62
  9. Distinctive:

    Use the current session's LLM to improve the prompt inline:

    line 88
  10. Distinctive:

    Stop. Do NOT execute.

    line 98
  11. Very specific:

    Create .env file in vault root (if not exists)

    line 209
  12. Very specific:

    Call track_event with event_name prompt_improved and properties:

    line 235

“Common” is measured: kept by at least 6 in 10 of the 10 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md
<!-- Generated from `.claude/skills/prompt-improver/SKILL.md` by `scripts/generate-agents-skills.py`. Do not edit. -->

Purpose

Transform vague, ambiguous prompts into rich, well-structured prompts. Uses Anthropic's prompt improvement capabilities when available, with graceful fallback to the current LLM.

How it works:

  1. User provides a vague prompt (e.g., "critique this doc")
  2. Skill improves the prompt using best available method
  3. Executes the improved prompt and returns results
  4. User sees the answer without seeing the improved prompt (unless flags are set)

Arguments

$PROMPT - The prompt to improve (required) $FEEDBACK - Optional feedback on what to improve (e.g., "Make it more detailed", "Add examples", "Focus on clarity") $TARGET_MODEL - Optional target model for the improved prompt (defaults to current model) $SYSTEM - Optional system prompt to improve alongside the user prompt


Qualifiers

Check if $PROMPT starts with a flag:

FlagBehavior
-pPrompt only - Show the improved prompt, don't execute
-vVerbose - Show the improved prompt, then execute
(none)Quick - Execute immediately without showing full prompt

Strip the flag from $PROMPT before processing.


Process

Step 1: Parse Flags and Extract Prompt

Check if $PROMPT starts with a flag (-p, -v) and extract:

  • mode: prompt-only, verbose, or quick (default)
  • original_prompt: The actual prompt text (flag removed)
  • feedback: Optional improvement guidance from $FEEDBACK

Step 2: Determine Improvement Method

Check in this order:

  1. Script available? Check if .scripts/improve-prompt.cjs exists
    • If yes → Use script (calls Anthropic API directly)
    • If no → Check for API key
  2. API key available? Check if ANTHROPIC_API_KEY is set in environment
    • If yes → Use Anthropic Messages API inline
    • If no → Fall back to current LLM

The fallback cascade:

Script (.scripts/improve-prompt.cjs)
    ↓ (if not available)
Anthropic Messages API (direct call)
    ↓ (if no API key)
Current LLM (Opus 4.5, Sonnet, etc.)

Step 3: Improve the Prompt

Method A: Script (preferred)

node .scripts/improve-prompt.cjs "$PROMPT" "$FEEDBACK" "$TARGET_MODEL" "$SYSTEM"

Method B: Anthropic Messages API (direct) Make API call with:

  • Model: claude-sonnet-4-5-20250929 (optimized for prompt engineering)
  • System Prompt: Prompt engineering expert persona (see below)
  • User Message: The original vague prompt
  • Temperature: 0.3

Method C: Current LLM Fallback Use the current session's LLM to improve the prompt inline:

  • Notify user: "💡 Using inline improvement (no API key configured). For best results, add ANTHROPIC_API_KEY to .env"
  • Apply the same prompt engineering system prompt
  • Continue with the improved result

Step 4: Handle Based on Mode

Mode: prompt-only (flag: -p):

  1. Show: > **Original:** [original_prompt]
  2. Show the enhanced_prompt in a code block
  3. Stop. Do NOT execute.

Mode: verbose (flag: -v):

  1. Show: > **Original:** [original_prompt]
  2. Show the enhanced_prompt in a collapsible block:
    <details>
    <summary>📝 Improved Prompt (click to expand)</summary>
    
    [enhanced_prompt]
    
    </details>
  3. Add --- separator
  4. Execute the enhanced_prompt and return results

Mode: quick (no flag - DEFAULT):

  1. Silently execute the enhanced_prompt
  2. Return results directly to user
  3. Do NOT show the improved prompt - user just sees the answer

Prompt Engineering System Prompt

Used for both API and fallback methods:

You are an expert prompt engineer trained in Anthropic's best practices. Your job is to transform vague, ambiguous prompts into clear, structured, effective prompts.

Analyze the user's prompt and improve it using these techniques:

1. **Structure**: Add clear sections with XML tags or markdown headers
2. **Clarity**: Be specific about format, length, and success criteria
3. **Context**: Include necessary background and define ambiguous terms
4. **Examples**: Add few-shot examples when helpful
5. **Chain of Thought**: For complex tasks, request step-by-step reasoning
6. **Constraints**: Make implicit constraints explicit

Return ONLY the improved prompt. Do not explain your changes or add meta-commentary.

{if $FEEDBACK exists: "Focus on: {$FEEDBACK}"}

Examples

/prompt-improver -p critique this strategy doc
→ Shows improved prompt only, doesn't execute

/prompt-improver -v critique this strategy doc
→ Shows improved prompt, then executes it

/prompt-improver critique this strategy doc
→ Just executes the improved prompt

/prompt-improver -v "review this code" "Focus on security issues"
→ Shows improved prompt focused on security, then executes

Improvement Template Reference

The improved prompt typically follows this structure:

# Task
[Clear statement of what to do]

# Context
[Background information needed]

# Instructions
1. [Step 1]
2. [Step 2]
3. [Step 3]

# Constraints
- [Constraint 1]
- [Constraint 2]

# Output Format
[Expected format and structure]

# Examples (if helpful)
[Input/output examples]

Error Handling

SituationBehavior
Script not foundFall back to API
No API keyFall back to current LLM with notification
API rate limitRetry with exponential backoff, then fall back
API errorFall back to current LLM
Network issuesFall back to current LLM

Key principle: The skill should NEVER fail completely. It always has the current LLM as ultimate fallback.


Setup (Optional)

For best results, add your Anthropic API key:

  1. Create .env file in vault root (if not exists)
  2. Add: ANTHROPIC_API_KEY=your-key-here
  3. Make the file owner-only: chmod 600 .env (API keys should never be readable by other users of the machine)

Without the API key, the skill still works using the current LLM session.


Philosophy

Meta-prompting: This skill uses Claude to improve prompts for Claude. It's prompt engineering as a service.

Invisible by default: The best tools disappear. Users ask naturally, get expert results, never see the complexity.

Progressive disclosure: Flags (-v, -p) let power users inspect and learn from the improvements.

Graceful degradation: Works everywhere - with full API access, partial access, or no external access at all.


Track Usage (Silent)

Call the mark_feature_used tool on the dex-analytics MCP server with prompt-improver. This ticks the feature's box in System/usage_log.md, which is what /dex-level-up reads to recommend features that have not been tried. It writes locally and sends nothing, so it runs whether or not analytics is enabled.

Analytics (Silent):

Call track_event with event_name prompt_improved and properties:

  • (no properties)

This only fires if the user has opted into analytics. No action needed if it returns "analytics_disabled".

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.