| name | promptify |
| description | Transform user requests into detailed, precise prompts for AI models. Use when users say 'promptify', 'promptify this', 'rewrite this prompt', 'make this prompt better/more specific', or explicitly request prompt engineering or improvement of their request for better AI responses. |
| allowed-tools | ["Bash","Read","Write","Edit","Grep","Glob","Agent","AskUserQuestion"] |
| metadata | {"category":"assistant","tags":["agent","prompt-engineering","clarity","specification","rewriting"],"status":"ready","version":10,"triggers":{"positive":["Promptify this: audit all skills against our findings doc.","Rewrite this prompt for better results.","Make this prompt more specific.","Can you improve this prompt?"],"negative":["Generate mock customer data in JSON format.","Write a React component for a login form.","Fix the bug in the checkout flow."]},"guidance":"`promptify` transforms user requests into precise, structured prompts.\nOnly activate when the user explicitly requests prompt improvement or rewriting.\nDo not activate for direct task execution requests."} |
Promptify
Transform user requests into detailed, precise prompts optimised for AI model consumption.
Core Task
Rewrite the user's request as a clear, specific, and complete prompt that guides an AI model to produce the desired output without ambiguity. Treat the output as specification language, not casual natural language.
Workflow
User-input rule: Any time this skill needs a decision, preference, or clarification from the user, it MUST use the AskUserQuestion tool with structured options — never free-text prose questions. This applies to the clarifying-questions path in step 2 and the delivery choice in step 7.
1. Analyze
Read the user's request carefully. Identify:
- The core intent and desired outcome
- Missing context (audience, domain, environment)
- Unstated constraints (length, tone, format)
- Expected output format
2. Decide Output Mode
Based on the analysis, choose how to proceed:
- Request is clear and complete — produce direct rewritten prompt
- 1-2 minor gaps — fill with marked
[Assumption: X] placeholders and proceed
- Major gaps (audience, scope, tech stack unknown) — ask clarifying questions via the
AskUserQuestion tool (structured options, not prose) before rewriting
- Multiple distinct objectives — split into separate prompts
3. Structure
Apply the four-block pattern to organise the prompt. See rules/structure-four-block-pattern.md.
- Context - Background, audience, domain
- Task - What the AI must do
- Constraints - Boundaries, rules, limitations
- Output Format - Exact structure of the response
Not every prompt needs all four blocks. Use only what adds clarity. For common prompt types, start from the skeletons in references/prompt-blueprints.md.
4. Draft
Apply the rules in rules/ to sharpen the prompt:
- Replace vague terms with measurable requirements
- Surface missing information as placeholders or questions (see
rules/clarity-surface-missing-info.md)
- Include success criteria defining what a good answer must include or avoid
- Mark any assumptions explicitly
- Add examples only where the desired output is genuinely ambiguous
- Specify exact format (headings, bullet style, length)
- Break complex tasks into numbered sequential steps
5. Self-Check
Before delivery, verify the draft against the rule checklist (see rules/quality-self-check.md):
If any check fails, fix the violation and re-check. Stop after the checklist passes or after two refinement passes (whichever comes first).
6. Output
Present the final prompt to the user as a markdown block, clearly labeled. Do not add commentary beyond the prompt itself.
7. Deliver
After presenting the prompt, use the AskUserQuestion tool (not a prose list) to ask the user how to proceed, offering these options:
- Execute now — Treat the generated prompt as your new instruction and proceed based on the current conversation context. Use your normal judgement to decide the best next action — plan complex tasks, implement simple ones directly, or ask clarifying questions if needed.
- Save to file — Write the prompt to a markdown file in the current working directory (e.g.
promptify-<timestamp>.md where <timestamp> is epoch seconds). Let the user know the file path.
Writing Guidelines
Structure
- Begin with a single short paragraph summarising the overall task
- Use headings (##, ###, ####) for sections only where appropriate (no first-level title)
- Use bold, italics, bullet points (
-), and numbered lists (1., 2.) liberally for organisation
- Never use emojis
- Never use
* for bullet points, always use -
Language
- Use plain, straightforward, precise language
- Avoid embellishments, niceties, or creative flourishes
- Think of language as specification/code, not natural language
- Be clear and specific in all instructions
Content
- Keep the prompt concise: 0.75X to 1.5X the length of the original request
- Do not add or invent information not present in the input
- Do not include unnecessary complexity or verbosity
Examples
Positive Trigger
User: "Promptify this: audit all skills against our findings doc."
Expected behavior: Use promptify guidance, follow its workflow, and return actionable output.
Non-Trigger
User: "Generate mock customer data in JSON format."
Expected behavior: Do not prioritize promptify; choose a more relevant skill or proceed without it.
Troubleshooting
Skill Does Not Trigger
- Error: The skill is not selected when expected.
- Cause: Request wording does not clearly match the description trigger conditions.
- Solution: Rephrase with explicit domain/task keywords from the description and retry.
Guidance Conflicts With Another Skill
- Error: Instructions from multiple skills conflict in one task.
- Cause: Overlapping scope across loaded skills.
- Solution: State which skill is authoritative for the current step and apply that workflow first.
Output Is Too Generic
- Error: Result lacks concrete, actionable detail.
- Cause: Task input omitted context, constraints, or target format.
- Solution: Add specific constraints (environment, scope, format, success criteria) and rerun.
