- Home
- Skills
- Prompt & Skill Engineering
- Prompt Clinic
Prompt Clinic
Turns a one-line request into a complete prompt: ten departments, a specialist clause set for the one you pick, a fixed output format and a pre-work question list. Runs offline, calls no model.
Free
Works with the AI Agents you already use
Prompt Clinic
Example session with this skill installed
I run maintenance on a filling line and I use ChatGPT to help me think through faults, but the
answers are always generic -- replace the sensor, check the wiring, call a technician. It does not
know my hardware and it does not ask. I have tried writing longer messages and it gets worse, not
better. I do not want to learn prompt engineering, I want to paste something that gets me a
troubleshooting order I can actually work through on the floor.
- Read your context and instructions
- Compiled the prompt clinic
That is the case this pack is built for, and the fix is structural rather than longer.
python compose.py --dept fault \
--obj "Siemens S7-1200 + Inovance IS620N" \
--sym "Servo trips F0021 overcurrent 3s after power-up, reset recovers it"
Standard output begins with the role, because the role is what stops the generic answer:
# Role
You are a field fault-diagnosis engineer with 15 years on PLCs, HMIs, VFDs and servos, who
asks about the symptom before theorising about the cause and never skips a verification step.
# My situation
- Problem type: Equipment / line fault diagnosis (Equipment Diagnostics)
- Object / equipment / system: Siemens S7-1200 + Inovance IS620N
- Specific symptom and request:
Servo trips F0021 overcurrent 3s after power-up, reset recovers it
# What I want you to do
Find the root cause of this fault and give me a troubleshooting sequence ordered as "verify
first, touch hardware later". Do not open with a parts-replacement list - I want to know which
layer is actually broken first.
Then the clause set that does the actual work. Clauses 1 to 6 are the general six; 7 to 10 are the
fault department's own, and those four are the sentences a generic answer never contains:
7. Rank candidate causes by **probability**, highest first, and for each one state how to
prove it true or false, and what tool that verification needs.
8. Separate the four layers - hardware fault, parameter configuration, program logic,
external conditions (power, air supply, noise, grounding) - and never blend them.
9. For anything live, pressurised or moving, write the stop / isolate / lock-out steps before
the operation steps.
10. If the symptom contradicts itself (an alarm code that does not match the observed
motion), call that contradiction out as its own item.
And it closes by refusing to guess about your machine:
# Before you start
If the information above is not enough for a reliable answer, confirm with me first in no more
than 5 questions. Do not invent the missing premises.
- Is there an alarm or fault code? Paste it verbatim, do not translate it.
- Is the fault permanent or intermittent? If intermittent, how often, and is there a pattern?
- Has anything changed in the last three days - program, parameters, wiring, a replaced part?
Paste that into the model you already use. The difference you will notice first is not better
prose, it is that the answer arrives as an ordered verification sequence instead of a parts list.
Connects securely to your tools. The creator never sees your data.
About this skill
A prompt usually fails before the model sees it. The problem, the object and the symptom were never separated, so the model answers the average of all three. "Servo keeps faulting" is not a request, it is a topic. The fix is not a bigger model, it is three named inputs and an explicit statement of what you do not want back.
This pack is the smallest tool that forces that separation. Pick one of ten departments, name the object, paste the symptom. It emits a finished prompt, not an answer.
THE TEN DEPARTMENTS
fault Equipment and line fault diagnosis -- layer separation, safety steps before work. code Code and script error triage -- root cause, then the minimum change. data Analysis and reporting -- metrics that can actually be computed, with definitions. copy Copywriting and marketing content -- a version worth testing, plus why it may fail. review Contract and technical document review -- clause by clause, ranked by exposure. decide Option comparison and selection -- criteria and weights fixed before scoring. learn Learning and training plan -- checkpoints, not a reading list. sell Client conversation and negotiation -- predict the refusal, then the reply. process Process and policy mapping -- find the real bottleneck, not a tidy document. other Something else -- the general clause set, and it asks rather than assuming.
Each department carries its own three or four clauses and its own output format. The fault department, for example, requires the four layers (hardware, parameters, program logic, external conditions) to stay separate, candidate causes ranked by probability with a verification method each, and stop / isolate / lock-out written before any live or pressurised operation. The copy department forbids a greeting and asks for the reason the piece might fail. That is the difference between ten prompts and one prompt with a noun swapped.
WHAT EVERY OUTPUT CONTAINS
Role A noun phrase, the years, and the habit that matters -- not "helpful
assistant".
My situation Your three inputs, verbatim. Nothing is paraphrased back at you.
What I want you One imperative sentence, department-specific.
Constraints Six general clauses, then three or four department clauses. The general six
are the anti-fluff set: no preamble; write "not sure" when unsure and say
what is missing; every conclusion traces to evidence; no "might", "perhaps",
"you may want to consider"; if my premise is wrong, say so first; density
over length.
Output format Four general rules, then a department format that wins on conflict.
Before you start Three department-specific questions. If the inputs are thin it asks instead
of inventing the premise.
The order is deliberate. Constraints come after the task so the model reads what to do before it reads what not to do, and the department format sits last because a format stated last is the one that survives a long context.
WHAT IT DOES NOT DO
It does not answer your question. It writes the prompt you paste into a model you already pay
for. There is no model call in this package and no API key anywhere.
It does not know your equipment, your plant or your numbers. Anything it cannot know becomes a
question in "Before you start", not a plausible-looking default.
It does not rank departments for you. If you are unsure, --dept other gives the general clause
set without a specialist frame; that is a valid choice, not a failure.
It does not remember previous runs. Each call is a pure function of its arguments.
REQUIREMENTS
Python 3.8 or newer, standard library only. One file, compose.py. Prints to stdout and writes
nothing; no network, no environment variables, no temporary files. python compose.py --list
prints the ten departments and what each one is for.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Prompt & Skill Engineering.
Generates prompts, not direct answers. Requires local Python execution. No persistent state or history between runs. Output must be manually pasted into a model.
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 buying a skill.
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 to use
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 --dept fault --obj "..." --sym "..."` 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, including its own: the entire template library is a string constant compiled into the file, every input arrives through argv, there is no path literal in the package and `os` is not imported. The prompt text does contain the word: the templates say "Who reads the conclusion, and what will they do differently" and "Do not hand me a rewritten file" -- English inside the text the tool generates, not a handle on anything. 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. There is no `open()`, no temp file, no cache and no log in the package. 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 template text mentions a URL (u.org.cn) as plain prose for a human reader; nothing requests it. Environment Variables -- off. No `os.environ`, no `getenv`, no `process.env`. `os` is not even imported.