- Home
- Skills
- Prompt & Skill Engineering
- Reverse Engineer Any Templates
Works with the AI Agents you already use
Reverse Engineer Any Templates
Give it one example's inputs and output, and get back the prompt that would have produced it: a reusable template with a variables section, slots, and the guesses marked
$5
/moSubscription · cancel anytime
Secure checkout via Stripe
- Always the latest version
- Delivered live to your agent via MCP
- Cancel anytime, access ends at period end
Reverse Engineer Any Templates
Example session with this skill installed
I have the inputs and the result, give me the prompt that made this.
Inputs: product = Trailblazer 40L hiking backpack; audience = weekend hikers; tone = upbeat; length = short
Output
Pack light, go far.
The Trailblazer 40L carries a weekend's worth of gear without weighing you down.
- Ventilated back panel keeps you cool on the climb
- Rain cover tucked in the base for sudden showers
- Hip-belt pockets for snacks and your phone
Ready for Saturday? Grab yours before the trail does.
- Read your context and instructions
- Compiled the reverse engineer any
Here's what I'm working from, so you can catch a mismatch before reading on:
- Inputs: product = Trailblazer 40L hiking backpack, audience = weekend hikers, tone = upbeat, length = short
Output: a bold four-word headline, one intro sentence that names the product, three feature bullets, and a closing question with a call to buy. About 50 words in total.
RECONSTRUCTED PROMPT
INPUTS
PRODUCT : [PRODUCT]
AUDIENCE : [AUDIENCE]
TONE : [TONE]
LENGTH : [LENGTH]
Write product marketing copy for {{PRODUCT}}, aimed at {{AUDIENCE}}, in a {{TONE}} tone. Keep the whole piece {{LENGTH}}. For short copy, stay at roughly 40 to 60 words in total.
Use this exact structure and Markdown formatting, with nothing before or after it:
1. A headline in bold. Keep it to five words or fewer, built from two short imperative phrases separated by a comma, and end it with a full stop. The headline names the core promise of the product, not the product itself.
2. One sentence on its own line that names the product in full and states its main benefit in terms of how {{AUDIENCE}} actually use it. Keep it under 15 words.
3. Exactly three bullet points, each starting with a hyphen. Each bullet names one concrete, plausible feature of {{PRODUCT}} first, then says in the same line what that feature does for {{AUDIENCE}} in a real moment of use. Keep each bullet under 10 words. Do not repeat a benefit across bullets.
4. A closing line made of a short question that ties to when {{AUDIENCE}} would next use the product, followed by a short imperative call to buy. Give the call to buy a light, playful twist.
Carry the {{TONE}} tone through word choice and rhythm: active verbs, second person, short sentences. Do not use exclamation marks, emojis, prices, discounts or superlatives such as "best". Do not add a title, a sign-off or any explanation of the copy.
MAPPING
PRODUCT (Trailblazer 40L hiking backpack): verbatim in part. The intro sentence uses "The Trailblazer 40L", and "hiking backpack" is transformed into the gear-and-trail vocabulary of the bullets and the closing line rather than being stated.
AUDIENCE (weekend hikers): transformed. It never appears word for word. It shows up as "a weekend's worth of gear", "on the climb", "Ready for Saturday?" and "before the trail does". It also decides which features were picked: ventilation, rain cover and snack pockets are things a day-to-weekend hiker cares about.
TONE (upbeat): transformed. It comes through in the imperative headline, second-person address, the light wordplay in the call to buy and quick sentence rhythm. Notably there are no exclamation marks, so upbeat is carried by energy, not punctuation.
LENGTH (short): structural. It shows as a total of about 50 words, a four-word headline, a single intro sentence, bullets of 7 to 8 words and one closing line.
ASSUMPTIONS
Observed directly in the example: the bold headline, one intro sentence naming the product, exactly three hyphen bullets in a feature-then-benefit order, a closing question followed by a call to buy, Markdown formatting, second-person address, and the total length of about 50 words.
Inferred, not confirmed:
- The 40 to 60 word range for "short" is a guess from this one output. The original may have used a different scale, or none at all.
- The per-line limits (headline of five words or fewer, intro under 15 words, bullets under 10 words) are set just above what the example shows. They are a plausible reading, not a known rule.
- The headline pattern of "two imperative phrases separated by a comma" may be a coincidence of this one headline rather than an instruction.
- The ban on exclamation marks, emojis, prices and superlatives is inferred from their absence. The original may simply not have produced them this time.
- The features were not among your inputs, so I've assumed the original prompt told the model to invent plausible features for the product. Another possibility is a hidden features input you didn't list. If you had one, add a FEATURES variable and have the bullets draw on it instead.
- I read the Markdown structure (bold headline, hyphen bullets) as something the prompt should reproduce, not as a one-off, because a prompt that drops it won't give you back this output.
- The prompt is written to work with any model, since you didn't name a target model.
This reconstruction rests on a single input-to-output pair, so some of these rules may be quirks of this one example rather than real instructions. Run the prompt on a fresh set of inputs, for example a different product, audience and tone. Or send me a second example from the same original prompt, and I can tighten the rules that are guesses today.
Subscription · always the latest version
- Secure checkout via Stripe
- Cancel anytime
- Delivered live via MCP
About this skill
For when you have a result you like and the values that went into it, but not the prompt. It works backward from one input-to-output pair: it separates what came from the inputs from what the instructions must have shaped (structure, length, tone, formatting), and writes the prompt that reproduces it on new inputs.
What it does
First it restates what it's working from, so a mismatch is caught early. Then it returns the reconstructed prompt with an inputs section declaring each variable and double-curly slots where they're used, plus explicit rules for every feature of the output's structure: headline shape, sentence count, bullet length, closing line. It marks which rules it observed and which it inferred, because one example can't tell a rule from a quirk, and it tells you how to firm them up with a second example.
Try it first
Paste the inputs and the output of a piece of copy you liked, and say "give me the prompt that made this". Run the prompt it returns on a different set of inputs.
What it needs
Both halves of at least one example: the input values and the output. If one is missing, it asks before reconstructing anything. No scripts, no network access and no API keys.
One example gives a template that may encode accidents of that example; a second or third pair makes it much more reliable, and it says which rules are guesses. It recovers a prompt that behaves like the original, not the original's wording. It doesn't run the prompt to check it; that test is yours. It isn't for image prompts.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 27 days ago
Permissions
Network Access is declared because the skill's text says "double curly" slots, which automated scans read as curl; the skill makes no network requests.