/stuck-or-pivot
For when an AI build, prompt or workflow isn't working. Separates what was expected from what actually happened and what was seen from what's guessed, runs one test, then helps decide whether to fix it, make it simpler or change the approach. Can also draft a clear help request to a colleague or forum.
How to call it
Once it's installed, ask for it by name, for example "use stuck-or-pivot on this"
or type /stuck-or-pivot
. Your AI tool also starts it without being asked when your request matches its description.
When it starts: someone says "it's not working", "I'm stuck", "it keeps doing X", "should I give up on this", or has tried the same fix twice.
Connections
Practises
Concepts it uses
Comes after
Try next
Used in blueprints
From the field, after this
The skill itself
This is the instruction text your AI tool reads, shown as written.
Stuck or pivot
Getting stuck is part of building. Help the person find out what’s wrong, then choose what to do next.
How this skill behaves
They describe what happened and give their own guess first; your guesses are labelled as guesses. “Just do it” gets your best diagnosis and a recommendation, with guesses marked.
Steps
- Expected versus actual. Ask, one at a time:
- What were you trying to do?
- What did you expect to happen?
- What actually happened?
- Does it repeat? Ask whether the same problem happens when they repeat the same steps with the same input.
- Seen versus guessed. Ask for their guess at the cause. Then sort the facts: what they saw, and what they’re guessing. Add your own possible causes, labelled as guesses.
- One test. Pick one change that would test the top guess, and say what result would change their mind. Run it, or have them run it, and compare.
- Choose:
- Fix it: the cause is clear and the fix is small.
- Make it simpler: cut scope or drop down a rung of the ladder, e.g. from agent to prompt.
- Change approach: the premise was wrong, or it wouldn’t make life easier even if it worked.
If the same kind of fix has failed twice, suggest testing the assumption behind it (
expert-council, attack-the-premise mode).
Ask for help (optional)
Draft a help request with them, in this shape:
I’m trying to [goal]. I asked the AI to [request]. I expected [X], but [what happened instead]. I tried [test]. My question is: [question].
Remove names, private data and anything confidential before they send it.
Output
A short decision note: goal, expected, actual, what was tested, the likely cause, the decision (fix, simplify or change) and why, and the next test.
Credit: Decision Note, ask-for-help format and build clinics from WDAI's Build Cycle (programs team), rewritten for the commons.
Licence: Apache-2.0 (instructions) and CC BY 4.0 (teaching text). From AI Practice Commons by Women Defining AI; method from the WDAI Build Cycle (beta).. Plain Markdown for agents: /skills/stuck-or-pivot.md