Works with the AI tools you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+22 moreClaude CodeCursorCodex CLIMuseDots
OpenClaw
Manus
Grok
Grok Bot+18 agentsPrompt Engineering Master
Convert vague AI goals into precise, testable, and robust prompt architectures using engineering principles and linting.
$7
Prompt Engineering Master
Example session with this skill installed
I have a prompt for extracting line items from invoices, but it keeps including the summary notes as products. How do I fix this? Mode: DEBUG.
- Read your context and instructions
- Compiled the prompt engineering master
- Generated the document
I have analyzed the provided scenario. The issue likely stems from a lack of clear data-instruction separation and missing negative constraints.
Prompt Audit & Debug
The original intent is to isolate structured line items from unstructured text. The failure occurs because the model conflates 'Summary Notes' with 'Line Items' due to proximity and shared terminology (e.g., prices).
| Diagnostic | Finding |
|---|---|
| Weakness | Undelimited input data allows "Summary" text to bleed into the "Extraction" logic. |
| Missing Component | Negative constraints (exclusion list) for non-product totals. |
| Instruction Conflict | A "General Extraction" instruction is too broad for specialized document parts. |
Improved Prompt Architecture
# TASK
Extract specific line items from the provided INVOICE DATA.
# DATA BOUNDARY
[START_INVOICE]
{{invoice_text}}
[END_INVOICE]
# EXTRACTION RULES
1. Only extract items with a specific quantity and unit price.
2. DO NOT extract "Total", "Tax", "Shipping", or "Summary Notes".
3. If no clear products are found, return an empty JSON array.
# OUTPUT CONTRACT
Return valid JSON only:
{
"items": [
{"description": "string", "qty": number, "price": number}
]
}
Key Changes:
Data Delimiters: Added [START_INVOICE] and [END_INVOICE] to prevent summary notes from being interpreted as instructions.
- Negative Constraints: Explicitly listed common summary fields to ignore.
- Empty State Policy: Defined behavior for cases where no valid items exist to prevent hallucinations.
Next steps
- Run the new prompt with an invoice containing long summary notes to verify separation.
- Add 2-3 "few-shot" examples of summary notes that should be ignored.
- Test with malformed OCR text to ensure the JSON structure remains valid.
prompt-engineering-master.pdf
PDF · document
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
What you get
About this skill
Vague prompts lead to inconsistent, hallucinated, or malformed AI outputs. This skill provides a systematic engineering framework to convert loose goals into high-precision instructions that actually work across different LLMs. It moves beyond simple "persona" prompting into architectural design, evidence-based debugging, and adversarial testing.
What it does
- Architectural Construction creates prompts using a rigorous loop of goals, context, requirements, and output contracts.
- Diagnostic Debugging analyzes existing prompts to identify why they fail and provides targeted refactors.
- Adversarial Red-Teaming stress-tests instructions against ambiguity, prompt injection, and malformed inputs.
- Workflow Decomposition breaks complex, multi-stage tasks into reliable chains or tool-aware agent instructions.
- Model Adaptation translates objectives into optimized formats for specific models like ChatGPT, Claude, or image generators.
How it works
- Selection Choose an operating mode such as BUILD, DEBUG, IMPROVE, or REDTEAM based on your current need.
- Analysis The agent reviews your goal or existing prompt against a library of reference architectures and failure patterns.
- Execution It runs deterministic Python scripts to lint the prompt text for vague adjectives, role inflation, or missing fallbacks.
- Delivery You receive a structured package including the final prompt, an explanation of the architecture, and a test matrix.
Frameworks & tools
Uses standard-library Python for deterministic linting and structural audits. The architecture follows established prompt engineering principles like few-shot learning, chain-of-thought design, and data-instruction separation.
Why this beats prompting it yourself
Most developers "guess and check" their way to a prompt that eventually breaks. This tool uses a repeatable engineering process and automated linting to catch edge cases, injection risks, and contradictory instructions before you ever hit the API.
Use cases
- Generating production-ready JSON schemas for LLM extraction tasks.
- Debugging why an agent is ignoring specific constraints or tools.
- Building a reusable prompt library for a team of developers.
- Securing prompts against user-input injection in public-facing apps.
Known limitations
This tool does not execute the target AI or provide live API testing; it performs structural and text-based engineering. It cannot guarantee deterministic results from non-deterministic models.
How to install
Works the same in every agent - Claude, Cursor, Codex, Copilot and 20+ more.
- 1
Download the ZIP
Free skills download straight away. Paid skills unlock right after purchase.
- 2
Unzip into your skills folder
Every agent reads skills from one folder on your machine. Drop the unzipped folder in there.
- 3
Ask your agent to use it
Restart the agent if it was already running. It picks the skill up automatically - no config needed.
Skills folder by agent
Click the path to copy it. Create the folder if it does not exist yet.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
30-day refund
Not a fit? Get your money back
Trust & safety
Security scanned
Verified clean 5 days ago
- Passed all security checks, Safe to install