detecting-weak-prompts
Skill0 starsChanged 35 days agoMITView on GitHub
Use when the user pastes text clearly addressed to an AI model — starts with "You are...", "Act as...", or gives direct imperative instructions to a model — and it's obviously missing role, constraints, or output format, even if they didn't ask for help improving it. Do not use for text addressed to a person, task requests directed at the assistant itself, code/config, or prompts that already look complete.
In this repository2 other files it gives its agents.
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 16
Only treat text as a candidate if it's addressed to a model, not a person or to you:
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Detecting Weak Prompts
Overview
Catches prompts that are clearly meant for an AI but are obviously missing a basic component, and offers once to strengthen them — without turning every pasted block of text into an interruption.
Core rule: offer exactly once per piece of text, then drop it regardless of the answer. Always answer whatever the user actually asked, independent of the offer.
Is this prompt text?
Only treat text as a candidate if it's addressed to a model, not a person or to you:
- Looks like this: "You are a...", "Act as...", "Your task is...", second-person imperative instructions written as if a model will execute them.
- Not this: a request directed at you ("summarize this," "fix this bug"), text addressed to a human ("tell your boss..."), code, config, or a document being discussed.
What counts as missing
Thin means it's clearly lacking at least one of: a defined role, explicit constraints (length/tone/scope), or a stated output format. Short isn't automatically thin — "Explain quantum computing in two sentences for a 10-year-old" already has a constraint and an implicit format; it doesn't need flagging.
Workflow
- Detect — apply the two checks above. If it's not AI-directed, or it's not obviously thin, do nothing — go straight to answering normally.
- Offer once — name the specific gap, one line: "That prompt doesn't specify a role or output format — want me to tighten it up?" Don't start drafting yet.
- Answer the actual request regardless of their answer to the offer — if they pasted the prompt for feedback, give feedback either way.
- On yes — hand off to the
prompt-builderskill to do the actual drafting. Don't duplicate its five-component workflow here. - On no, or no response to the offer — drop it. Never re-offer on that same text again, even if it comes back up later in the conversation.
Common mistakes
| Mistake | Fix |
| Flagging a prompt addressed to a person ("tell him to clean the garage") | Check who it's addressed to before flagging — only AI-directed text counts. |
| Flagging an already-complete prompt because it's short | Thin means missing a component, not low word count. |
| Re-offering after a decline | One offer per piece of text, ever. |
| Treating "summarize this document" as a weak prompt to fix | That's a task request directed at you, not a prompt draft — just do the task. |
| Drafting the improved prompt yourself instead of handing off | This skill only detects and offers; prompt-builder does the drafting. |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.