---

name: metaprompt-factory description: This skill should be used when generating structured metaprompts for repeatable tasks. Triggers include "create a metaprompt for X", "generate a review prompt", "make a prompt template", or when building prompts for cursor-agent, Codex, or other AI tools.

metaprompt-factory

generate structured XML metaprompts for repeatable AI tasks. adapts to domain, enforces quality gates, cites sources.

philosophy

"a metaprompt is a reusable product spec for an AI worker"

principle	application
make implicit explicit	encode assumptions, boundaries, constraints
separate policy from procedure	principles in philosophy, steps in workflow
optimize for repeatability	generalize across similar tasks
source-grounded claims	cite official docs for technical claims
no phantom specifics	don't invent APIs, versions, paths

when to use

use	skip
repeatable AI tasks	one-off questions
cursor-agent prompts	simple commands
Codex review prompts	ad-hoc exploration
PR audit templates	conversational tasks
domain-specific workflows	research queries

workflow

1. intake: task description, domain, constraints
2. classify: determine domain (coding, review, ops, research)
3. structure: apply metaprompt template from factory reference
4. parameterize: extract variables, set defaults
5. validate: check against rubric (clarity, completeness, groundedness)
6. output: XML metaprompt ready for use

input packet

minimal viable inputs:

• target_task: what the prompt should accomplish
• domain: coding | review | ops | research
• outputs: expected deliverables

optional:

• constraints.must_do: required behaviors
• constraints.must_not_do: forbidden behaviors
• tooling: available tools (cursor-agent, Codex, etc.)
• citations.required: whether to enforce citations

output format

<?xml version="1.0" encoding="UTF-8"?>
<metaprompt name="{name}" version="1.0" updated="{date}">
  <philosophy>...</philosophy>
  <parameters>...</parameters>
  <instructions>...</instructions>
  <workflow>...</workflow>
  <output_requirements>...</output_requirements>
  <failure_modes>...</failure_modes>
</metaprompt>

quality rubric

dimension	checks
clarity	roles explicit, I/O unambiguous
completeness	workflow + failure modes + output spec
groundedness	technical claims cited or marked "proposed"
reusability	parameters cover variability
operability	machine-readable summary exists

references

• references/factory-template.xml - base metaprompt template
• references/domain-modules.md - coding, review, ops modules
• references/rubric.md - validation checklist

anti-patterns

• phantom specifics: inventing APIs, versions, paths
• policy bloat: too many rules, loses clarity
• missing failure modes: what happens when things break?
• vague outputs: "generate a response" vs structured spec