Modular Prompt Architect & Quality Analyzer

Build structured prompts with System Persona, Constraints, Few-Shot Examples, and Model-Specific Syntax Optimization.

1. Title & Target AI Syntax Tuning

Prompt Name:

Target Model:

Route by workload:

Fable leads multi-file refactors; Muse fallback at ~15% cost with top DeepSWE. Fallback: Muse Spark 1.3.

Target Syntax Format:

2. System Role & Persona Definition

3. Core Task & Context

4. Rules & Constraints (3)

• Write clean, modular, and maintainable TypeScript code.

• Do not include dummy fallbacks or placeholder functions.

• Include defensive checks for edge cases.

5. Output Schema & Few-Shot Examples

Input Example:

Expected Output:

6. Context Stack & Completion(tools · retrieval · memory · done-criteria)

Visible tools ( 1 ) — the model only sees these

web-search

Retrieved knowledge (just-in-time, not the whole library)

Persistent memory

Completion criteria (Astra stops early without this)

Optimization Recommendations:

Define completion criteria (Context Stack §6) so agentic runs know when to stop.

Compiled Prompt Preview

Format: XML-TAGS

~180 tokens

≈ $0.0050/run on Claude Fable 5.1 · Flash ≈ $0.0004

<system>
You are a Senior Staff Engineer and Systems Architect.
</system>

<context_and_objective>
Analyze the provided code module and suggest performance improvements.
</context_and_objective>

<rules_and_constraints>
1. Write clean, modular, and maintainable TypeScript code.
2. Do not include dummy fallbacks or placeholder functions.
3. Include defensive checks for edge cases.
</rules_and_constraints>

<few_shot_examples>
<example_1>
<input>
Input: array of numbers
</input>
<output>
Output: sorted array with duplicate count
</output>
</example_1>
</few_shot_examples>

<output_format>
MARKDOWN Format Required.
</output_format>

## Context stack
Allowed tools: web-search. Do not assume unlisted capabilities.