Template Engineering
Guide for creating meta-prompt templates - prompts that build prompts.
When to Use
-
Creating new slash commands that generate plans
-
Encoding team workflows into reusable templates
-
Designing meta-prompts for chores, bugs, features
-
Analyzing existing templates for improvement
Template Anatomy
Every effective template has five sections:
- Purpose Section
Clear description at the top:
[Work Type] Planning
Create a new plan in specs/*.md to resolve the [Work Type] using the exact specified markdown Plan Format.
- Instructions Section
Guidance for the agent:
Instructions
- You are writing a plan, not implementing
- Use your reasoning model: THINK HARD about the plan
- Focus on files in the Relevant Files section
- Include validation commands that verify completion
- Fill in every section of the Plan Format
- Relevant Files Section
Guide context loading:
Relevant Files
Focus on the following files:
- README.md (project context)
- src/** (source code)
- tests/** (test patterns)
- docs/** (documentation)
- Plan Format Section
Template with placeholders:
Plan Format
[WorkType]: <name>
Description
<what needs to be done>
Relevant Files
<files to modify>
Step by Step Tasks
<numbered actionable tasks>
Validation Commands
<commands proving completion>
Notes
<edge cases and considerations>
- Parameter Section
Accept user input:
[WorkType]
$ARGUMENTS
Design Workflow
Step 1: Identify the Work Type
What class of problems does this template solve?
-
Chores: Maintenance, updates, cleanup
-
Bugs: Investigation and fixes
-
Features: New functionality
-
Refactors: Code improvement
Step 2: Define the Plan Structure
What sections does this work type need?
Work Type Essential Sections
Chore Description, Files, Tasks, Validation
Bug Problem, Solution, Reproduce, Root Cause, Tasks, Validation
Feature User Story, Implementation Plan, Testing, Acceptance
Refactor Before/After, Migration, Rollback
Step 3: Write Instructions
Guide the agent's reasoning:
-
What should they focus on?
-
What reasoning depth is needed? ("think hard")
-
What common mistakes should they avoid?
-
What format requirements are strict?
Step 4: Specify Relevant Files
Limit scope to improve quality:
-
Core files for this work type
-
Test files to understand patterns
-
Config files if relevant
-
Avoid overly broad patterns
Step 5: Test with Examples
Validate your template:
-
Run with simple input
-
Check generated plan completeness
-
Try edge cases
-
Refine based on results
Quality Checklist
Before deploying a template:
-
Purpose clearly states what template does
-
Instructions include "THINK HARD" for reasoning
-
Relevant Files are specific, not too broad
-
Plan Format has all necessary sections
-
Validation Commands section is mandatory
-
Parameter uses $ARGUMENTS correctly
-
Output location specified (specs/*.md)
Common Patterns
Reasoning Activation
Use your reasoning model: THINK HARD about the plan and the steps to accomplish this work.
File Output Pattern
Create a new plan in specs/*.md using the following naming: specs/[worktype]-[descriptive-name].md
Validation Mandate
The Validation Commands section is REQUIRED. Every plan must include commands that prove the work is complete.
Anti-Patterns to Avoid
Too Broad Relevant Files
Bad: */ (everything) Good: src/auth/, tests/auth/ (focused)
Missing Validation Section
Bad: No way to verify completion Good: Specific test and lint commands
Vague Instructions
Bad: "Make a good plan" Good: "Focus on error handling edge cases, include rollback steps"
No Output Location
Bad: Plan written somewhere undefined Good: specs/[worktype]-[name].md pattern
Related Memory Files
-
@template-engineering.md - Core concepts and examples
-
@meta-prompt-patterns.md - Prompt hierarchy
-
@plan-format-guide.md - Standard plan structures
Version History
- v1.0.0 (2025-12-26): Initial release
Last Updated
Date: 2025-12-26 Model: claude-opus-4-5-20251101