← Back to Skills Marketplace
176
Downloads
2
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install covert-native-language-to-ai-firendly-prompt
Description
Transforms casual or voice-transcribed user requests into precise, AI-optimized prompts. Handles mixed languages, vague input, and ambiguity. Reduces task ex...
README (SKILL.md)
Prompt Refiner
Turn messy input into structured, AI-optimized prompts on the first try.
When to Use
- Voice transcription input (speech-to-text)
- Casual, informal, or mixed-language requests (English + Chinese)
- Vague or ambiguous requests (missing target, unclear scope)
- Complex multi-step tasks that benefit from chaining
- Before destructive actions (delete, restart, deploy)
Skip if: request is already specific, task is simple/low-stakes, or user says "just do it."
Core Framework: TCREI
Google's prompt engineering framework — apply to every refined prompt:
| Component | What to include |
|---|---|
| Task | Action verb + specific target. "Summarize the sales report for Q1" |
| Context | Background, environment, constraints. "Account: [email protected], NZ timezone" |
| References | Examples, templates, tone samples. "Match this format: [example]" |
| Evaluate | How to judge the output. "Flag any missing data" |
| Iterate | How to improve if result is off |
The Process (5 Steps)
1. Analyze
Identify: Intent · Target · Constraints · Gaps · Language
2. Assign Persona (Always)
Give the AI a role that matches the task:
- Code task →
"You are a senior Node.js engineer" - Email task →
"You are a professional business writer" - Data task →
"You are a data analyst specializing in sales metrics" - Security task →
"You are a cybersecurity expert reviewing for vulnerabilities"
3. Clarify (If Critical Gaps Exist)
Ask ONE focused question — not multiple.
- ✅ "Which file —
api/validate.jsorapi/auth.js?" - ❌ "Which file? What language? What to check? When is the deadline?"
4. Construct the Structured Prompt
Persona: [Role + expertise relevant to the task]
Task: [Action verb + specific target]
Context: [System, environment, account, paths, dates]
References: [Examples, templates, or few-shot samples when format matters]
Requirements: [Constraints, scope, edge cases, what NOT to do]
Output: [Format, destination, success criteria, level of detail]
Advanced techniques — apply when appropriate:
- Few-shot: Add 1–2 input/output examples when format consistency matters
- Chain of Thought: Add
"Think step by step:"for complex reasoning - Prompt Chaining: Break multi-step tasks into linked sub-prompts
- Meta Prompting: Ask AI to refine the prompt itself before executing
See references/techniques.md for when/how to use each technique.
5. Confirm & Execute
- Destructive/complex actions: Show 1-sentence summary → get confirmation
- Safe/obvious tasks: Execute directly
Quick Checklist
Before executing, verify:
- ✅ Persona assigned
- ✅ Intent is clear (specific action + target)
- ✅ Context is concrete (real paths, accounts, dates)
- ✅ Requirements are testable
- ✅ Output format defined
- ✅ Success criteria stated
Real Examples
See references/examples.md for complete worked examples including:
- Voice transcription (Chinese) → Gmail check
- Vague code review → structured debug prompt
- Mixed-language service restart
- Complex multi-step task with chaining
Common Anti-Patterns to Avoid
| Anti-Pattern | Fix |
|---|---|
| Too many requirements in one prompt | Split into chained sub-prompts |
| Vague success criteria ("write a good report") | Define measurable criteria |
| No edge case handling | Add: "If X is missing, do Y" |
| Tweaking temperature instead of the prompt | Improve prompt structure first |
| Negative instructions only ("don't do X") | Tell it what TO do instead |
Usage Guidance
This skill appears to do what it says and contains only a small local CLI script. Before you install or use it: (1) do not paste secrets (passwords, API keys, tokens) into prompts; (2) be cautious when the skill asks for account names, file paths, or service names—only provide the minimum context needed; (3) require explicit confirmation before allowing the agent to execute any destructive or system-level actions referenced in a refined prompt; (4) if you plan to run the included CLI, review scripts/refine.js locally (it contains no network calls) and run it in a controlled environment; (5) if you want to avoid accidental execution, consider disabling autonomous invocation or instruct the agent to never execute commands automatically. If the skill were to add network calls, remote installs, or required credentials later, reassess (that would raise the risk).
Capability Analysis
Type: OpenClaw Skill
Name: covert-native-language-to-ai-firendly-prompt
Version: 1.0.1
The 'prompt-refiner' skill is a legitimate utility designed to transform informal or vague user requests into structured, AI-optimized prompts. The core logic in 'scripts/refine.js' is a safe text-processing script that uses regular expressions to analyze intent and clarity without any network or file system side effects. The instructions in 'SKILL.md' and the examples in the 'references/' directory are well-documented and align strictly with the stated purpose of improving prompt engineering workflows.
Capability Assessment
Purpose & Capability
Name/description (prompt refinement for messy/multi-language input) matches SKILL.md, examples, and the included CLI script (scripts/refine.js) which performs language/intent detection and builds structured prompts. No unrelated binaries, env vars, or external services are required.
Instruction Scope
SKILL.md confines the skill to analysis, clarification, persona assignment, and constructing structured prompts. However it also includes guidance to 'Confirm & Execute' and examples that embed system-specific context (email accounts, file paths, launchd services, health checks). The skill itself does not include code to perform those destructive actions, but the instructions could encourage an agent to ask for or accept sensitive operational details and then trigger execution via other capabilities. Recommend ensuring the agent does not automatically execute system-level actions without explicit user confirmation.
Install Mechanism
No install spec; this is instruction-first with a small local CLI JS file. There are no downloads, remote installers, or extracted archives in the skill package. package.json and scripts/refine.js are local and self-contained.
Credentials
The skill declares no required env vars, credentials, or config paths. Still, SKILL.md and examples explicitly encourage including account identifiers, system paths, and service names in the constructed prompts — which may lead users or agents to provide sensitive information (email accounts, local file paths, service names). This is proportionate to prompt-refinement functionality but users should avoid supplying secrets (passwords, tokens) and be cautious about automatic execution of actions that use that context.
Persistence & Privilege
Skill does not request persistent presence (always:false). It does not modify other skills or system settings. Autonomous invocation is allowed (platform default) but not combined with other privilege escalation indicators in this package.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install covert-native-language-to-ai-firendly-prompt - After installation, invoke the skill by name or use
/covert-native-language-to-ai-firendly-prompt - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.1
**Expanded prompt engineering and best practices integrated.**
- Added prompt engineering best practices, including persona assignment, few-shot examples, chain of thought, and prompt chaining.
- Introduced the Google TCREI framework for structured prompt design.
- Now explicitly assigns AI personas relevant to each task type.
- Includes a quick checklist to ensure prompt clarity, context, requirements, and output format.
- Added references/examples.md and references/techniques.md for in-depth techniques and sample prompts.
- Expanded guidance for handling advanced and multi-step tasks, including anti-patterns to avoid.
v1.0.0
Turn Messy Input into powerful AI Prompts
Metadata
Frequently Asked Questions
What is Prompt Refiner?
Transforms casual or voice-transcribed user requests into precise, AI-optimized prompts. Handles mixed languages, vague input, and ambiguity. Reduces task ex... It is an AI Agent Skill for Claude Code / OpenClaw, with 176 downloads so far.
How do I install Prompt Refiner?
Run "/install covert-native-language-to-ai-firendly-prompt" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Prompt Refiner free?
Yes, Prompt Refiner is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Prompt Refiner support?
Prompt Refiner is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Prompt Refiner?
It is built and maintained by jamesxu81 (@jamesxu81); the current version is v1.0.1.
More Skills