- Home
- Skills
- Prompt & Skill Engineering
- Prompt Expander
Prompt Expander
One plain sentence in, a prompt with role, method, constraints and output format out. It classifies the sentence into one of nine task types, then names the information it left out as questions to answer first.
Free
Works with the AI Agents you already use
Prompt Expander
Example session with this skill installed
I know what I want to ask but it comes out as one long sentence, and the reply comes back on the
wrong level -- I get a lecture instead of a plan. Someone told me to "add context" so I added three
paragraphs and now it answers a slightly different question. I want to write one sentence the way
I would say it to a colleague and get back something the model can actually execute.
- Read your context and instructions
- Compiled the prompt expander
That is exactly the input shape this pack expects -- one sentence, quoted back, no rewriting.
python compose.py --say "My boss wants a report next week on why this line keeps stopping and
I do not know where to start"
It classified that as analyze, and the output opens by locking the sentence so the model cannot
drift to a tidier version of it:
# Role
You are an analyst who defines the question before touching the data, and objects to metric
dumping.
# The original request (the one sentence I wrote)
> My boss wants a report next week on why this line keeps stopping and I do not know where to
start
# Task
Treat that sentence as a formal brief and carry it out. Do not comment on how well it was
written. Just do it.
# Method
1. Translate the business question into a computation: numerator, denominator, time window,
de-duplication rule.
2. For each metric, state what it cannot answer.
3. Name the three places the data most likely goes wrong.
It closes with the part that makes the difference. The sentence said "next week", so the deadline
probe found its match; the other three did not, and each one becomes a question:
# Before you start
My sentence is missing the following. **Ask first, do not assume**; if I cannot answer,
continue with the default you state openly as an assumption:
- What is the budget or cost ceiling, and what happens if it is exceeded?
- What have you already tried, and what happened?
- Are there hard lines - compliance, competitors, wording - that must not be crossed?
Add "we already swapped the sensor and the fault moved to another axis" and the answer changes
shape, because the model no longer spends its first paragraph on the thing you already did. That is
the return on this tool: one sentence in, one prompt out, and the gaps named instead of guessed.
Connects securely to your tools. The creator never sees your data.
About this skill
An expander that only makes text longer is worthless. The value is in the gap analysis: a sentence carries a want and usually drops the reader, the deadline, the budget, the thing already tried and the hard lines. A model given the sentence alone fills those five gaps with plausible defaults, and plausible defaults are why the answer comes back confident and wrong.
This pack makes the gaps visible. One argument, --say, in plain English. It classifies the
sentence into one of nine task types, assembles the prompt for that type, and prints the questions
your sentence did not answer under "Before you start" -- ask first, do not assume.
THE NINE TASK TYPES
diagnose something is broken and you want the cause create you want content made analyze you want a number, a comparison or an attribution decide you want a choice made defensible learn you want a subject taught in an order build you want something implemented audit you want risk or defects found in something that exists persuade you want someone to agree or act general none of the above; the general frame without a specialist method
Classification is keyword-based and runs on the sentence you typed, not on a model. That is a deliberate limit, not an optimisation: it means the tool works offline and returns the same answer twice. The cost is that a sentence whose vocabulary belongs to two types gets filed under the nearest one, and the method section will be for the neighbouring intent. The role, the constraints and the output format are type-independent, so the prompt is still usable -- but if the method looks wrong, rephrase the sentence around its verb and run it again.
THE FIVE PROBES
Every sentence is checked for five things, each with its own keyword list. A probe that finds no match becomes a question at the end of the prompt:
reader Who reads or uses the result? Their role, and what they care about. deadline When is it needed, and how many hours can go into it? budget What is the cost ceiling, and what happens if it is exceeded? history What have you already tried, and what happened? limits Hard lines -- compliance, competitors, wording -- that must not be crossed.
WHAT EVERY OUTPUT CONTAINS
Role Chosen for the task type, phrased as a habit rather than a job title.
The original Your sentence quoted back once and locked. Not paraphrased, not improved.
Task Treat the sentence as a formal brief and carry it out; do not comment on how
well it was written.
Method Three ordered steps for the task type.
Constraints The same six general clauses as the other packs in this library.
Output format Four general rules, then a task-specific one that wins on conflict.
Before you start The probes your sentence failed, phrased as questions.
WHAT IT DOES NOT DO
It does not answer your question. It hands you the prompt; the answering happens in whatever model you paste it into. It does not rewrite your sentence to sound professional. The sentence is quoted verbatim and used as the brief -- improving the wording would hide the gaps this tool exists to expose. It does not invent the missing information. Every gap becomes a question with an explicit "ask first, do not assume" instruction attached. It does not store anything. No history, no profile, no memory between runs.
REQUIREMENTS
Python 3.8 or newer, standard library only. One file, compose.py. Prints to stdout and writes
nothing. python compose.py --list prints the nine task types.
Does not generate the final answer, only the prompt. No state is saved between runs. Requires manual data entry for variables identified as missing.
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 21 days ago
- Free with an Agensi account
Permissions
Terminal -- usage, not behaviour. This is a script skill: the operator runs `python compose.py --say "..."` in their own shell and reads the prompt from stdout. Nothing in the package starts a process, opens a shell, evaluates a string or downloads anything; `compose.py` imports argparse, json and sys and nothing else. Read Files -- off, and correctly off. `compose.py` opens no path at all: the nine task types, the five probes and the clause sets are string constants inside the file, the sentence arrives through argv, there is no path literal and `os` is not imported. The word is there in the text, though: the probe set is written as "Fix who the reader is and what state they are in", and one of the five questions the tool asks back is "Who reads or uses the result?" -- product prose, not a call. Verify in one line: `grep -n "open(" compose.py` returns nothing. The submit form's own auto-detect produced Terminal and nothing else on this ZIP, which is the same reading. Write Files -- off. The only output is stdout -- no `open()`, no temp file, no cache, no history file. Nothing is remembered between runs. Browser -- off. Nothing opens a page, drives a browser or imports a browser library. Network Access -- off. No socket, no HTTP client, no DNS, no telemetry, no update check. The u.org.cn URL in the prose is text for a human reader; nothing requests it. Environment Variables -- off. No `os.environ`, no `getenv`, no `process.env`; `os` is not imported.