agentleFS
kanwia-aikanwia-ai/claude-skills/skills/prompt-polisher/SKILL.md

prompt-polisher

Skill0 starsChanged 12 months agoMITView on GitHub

Use when receiving messy, unstructured input like voice transcriptions, stream-of-consciousness notes, or rough document content that needs to be transformed into a polished, optimized prompt. Cleans up filler words, extracts intent, asks clarifying questions, applies Claude 4.x/Opus 4.5/Sonnet 4.5 best practices, and previews the polished prompt for approval before execution. Trigger phrases include "polish this", "clean this up", "turn this into a prompt", or when input is clearly rough/unstructured.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. cnfjlhj/ai-collab-playbook★ 456same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 13 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈13

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the input appears to be voice transcription or casual speech, clean it up:

    line 25
  2. Distinctive:

    Wait for user response before proceeding.

    line 72
  3. Distinctive:

    Add contextual framing (explain WHY, not just WHAT)

    line 80
  4. Distinctive:

    Use positive framing ("use prose" not "don't use bullets")

    line 81
  5. Distinctive:

    Use action-oriented language ("Change this" not "Can you suggest")

    line 83
  6. Very specific:

    Use normal tone (avoid aggressive CAPS/MUST/CRITICAL)

    line 86
  7. Distinctive:

    Add anti-over-engineering guardrail: "Keep solutions simple.

    line 89
  8. Distinctive:

    Avoid the word "think" → use "consider", "evaluate", "believe"

    line 90
  9. Distinctive:

    Keep prompts token-efficient (don't over-pad)

    line 92
  10. Distinctive:

    Add more explicit step-by-step guidance for intricate tasks

    line 98
  11. Distinctive:

    Keep each section focused and concise

    line 127
  12. Distinctive:

    Use bullet points only when listing multiple items

    line 128
  13. Distinctive:

    Always preview before executing - never skip the approval step

    line 162
  14. Distinctive:

    If the input is already well-structured, acknowledge this and offer minor polish only

    line 163

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “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

Prompt Polisher

Transform messy, unstructured input into polished, Claude-optimized prompts using Anthropic's best practices.

When to Activate

Activate this skill when:

  • User provides voice transcription content (contains filler words, self-corrections, casual speech)
  • User dumps stream-of-consciousness notes
  • User pastes rough document content that needs structuring
  • User explicitly asks to "polish", "clean up", or "turn this into a prompt"
  • Input is clearly unstructured and intended as task instructions

Workflow

Execute these stages in order:

Stage 1: Voice Cleanup

If the input appears to be voice transcription or casual speech, clean it up:

Remove:

  • Filler words: "um", "uh", "like", "you know", "basically", "actually", "literally", "right?"
  • Self-corrections: "no wait", "I mean", "actually scratch that", "let me rephrase"
  • False starts: repeated or abandoned sentence beginnings
  • Verbal tics: "okay so", "let me think", "hmm"

Normalize:

  • Convert spoken patterns to written form
  • Fix run-on sentences
  • Add proper punctuation

Stage 2: Intent Extraction

Parse the cleaned input to identify:

ElementWhat to Find
Core taskWhat does the user actually want done?
Referenced filesDocuments, project files, URLs mentioned
ConstraintsLimitations, requirements, boundaries
PreferencesStyle, format, approach preferences
Success criteriaWhat does "done" look like?
ScopeSingle task vs. multi-step project

Stage 3: Gap Detection

Scan for missing or ambiguous elements:

  • Ambiguous requirements ("process them somehow")
  • Missing success criteria (no clear definition of "done")
  • Unclear scope (single task or multi-step?)
  • Format unspecified (code? prose? structured output?)
  • Model unknown (Opus 4.5 or Sonnet 4.5?)
  • Technical choices undefined (which library/approach/pattern?)

If gaps exist, ask ALL clarifying questions at once:

Before I polish this prompt, I need to clarify a few things:

1. [First gap-specific question]
2. [Second gap-specific question]
3. Which Claude model are you using? (Opus 4.5 / Sonnet 4.5 / Not sure)

Wait for user response before proceeding.

Stage 4: Apply Best Practices

Reference reference.md for detailed guidelines. Apply these transformations:

Universal (Claude 4.x):

  • Make instructions explicit and direct (no hints)
  • Add contextual framing (explain WHY, not just WHAT)
  • Use positive framing ("use prose" not "don't use bullets")
  • Structure with XML tags for complex inputs
  • Use action-oriented language ("Change this" not "Can you suggest")
  • Include chain of thought prompts for complex reasoning
  • Define clear success criteria
  • Use normal tone (avoid aggressive CAPS/MUST/CRITICAL)

If Opus 4.5:

  • Add anti-over-engineering guardrail: "Keep solutions simple. Only make changes directly requested."
  • Avoid the word "think" → use "consider", "evaluate", "believe"
  • Leverage planning strength: "Build an editable plan before executing"
  • Keep prompts token-efficient (don't over-pad)
  • Provide policy/constraint context when relevant

If Sonnet 4.5:

  • Break complex tasks into explicit steps
  • Optimize for speed/throughput
  • Add more explicit step-by-step guidance for intricate tasks

Stage 5: Structure the Prompt

Generate the polished prompt using this template:

<context>
[Background info, relevant project context, referenced files, WHY this task matters]
</context>

<task>
[Clear, explicit statement of what needs to be done - action-oriented]
</task>

<requirements>
[Specific constraints, boundaries, must-haves, technical choices]
</requirements>

<success_criteria>
[What "done" looks like, how to verify completion]
</success_criteria>

<approach>  <!-- Optional: include for complex tasks -->
[Suggested approach OR "Build an editable plan before executing"]
</approach>

Formatting rules:

  • Keep each section focused and concise
  • Use bullet points only when listing multiple items
  • Omit sections that aren't relevant to the task
  • For simple tasks, a streamlined version without XML tags is acceptable

Stage 6: Preview for Approval

Present the polished prompt to the user:

Here's your polished prompt:

---

[POLISHED PROMPT CONTENT]

---

**Options:**
- **Execute** - Run this prompt now
- **Edit** - I'll modify something first
- **Redo** - Start over with different clarifications

Wait for user approval before executing.

Stage 7: Execute or Iterate

Based on user choice:

  • Execute: Immediately act on the polished prompt as if the user had typed it directly
  • Edit: User modifies the prompt, then re-preview
  • Redo: Return to Stage 1 with new input or Stage 3 for new clarifications

Important Notes

  • Always preview before executing - never skip the approval step
  • If the input is already well-structured, acknowledge this and offer minor polish only
  • Learn from user feedback - if they frequently edit a certain way, note the pattern
  • For very short/simple inputs, use a streamlined format without full XML structure
  • The goal is USEFUL prompts, not LONG prompts - be concise

See Also

  • reference.md - Detailed Anthropic best practices for Claude 4.x, Opus 4.5, and Sonnet 4.5
  • examples.md - Before/after transformation examples

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.