Prompt Architecture
prompt architecture Link to heading
External validation: Passed the Nate B. Jones 5-Level Prompting assessment with a 5/5 rating covering all critical competencies from basic prompt construction to meta-prompting systems design.
Advanced prompt architecture Link to heading
Multi-pattern integration — Research Brief, Few-Shot, Action-Oriented, and Constitutional patterns combined in real-world scenarios, not toy examples.
Three-level specificity mastery
- Task Specificity — precise deliverable definition
- Context Specificity — comprehensive background framing
- Output Specificity — detailed format and structure requirements
Context management & token efficiency Link to heading
- Progressive disclosure — 4-step strategies for large document processing with adaptive chunking and user calibration
- Personalization integration — CV-based filtering systems that customize AI interactions to user expertise gaps
- Interactive optimization — feedback loops for real-time prompt refinement and validation
Meta-prompting & systems design Link to heading
- Quality-assurance meta-prompts — pre-execution analysis with ambiguity detection, context validation, parameter management
- Model optimization — frameworks for adapting prompts to specific model strengths (speed, reasoning, external search)
- Anticipatory planning — systems that predict and prepare for follow-up questions and edge cases
Prompt-as-code Link to heading
The prompt library is versioned with code, not pasted into chat windows.
- 377-line XML prompt schema (
ideation-intent/prompt-schema.md) separating model-native capabilities (toolset, memory, provenance) from user-defined orchestration (abstract, role, guardrails) - 8 promptsets under
prompts/— Dark Code audit/context/comprehension, Proper Skills deployment, etc. - 173+ lines of interactive session logs documenting iterative prompting
Schema-first governance for agent round-trip Link to heading
The same discipline applies to ontology. Custom SHACL 1.2 Turtle vocabulary (mdu: namespace, sh:sparql constraints with $data substitution into DuckDB SQL) is designed specifically so an agent can round-trip the Turtle without losing dialect detail. From one .ttl file: Postgres / BigQuery / Databricks DDL plus Parquet / Iceberg (S3 + AWS Glue) / DuckLake / Nimtable lake schemas.
Key strengths Link to heading
- Systems thinking — prompting as workflow design, not tactical instruction improvement
- Pattern recognition — intuitive combination of multiple prompting patterns
- Efficiency optimization — comprehensiveness balanced against token economy
- Human-in-the-loop design — interactive, adaptive systems
- Knowledge architecture — frameworks for scalable skill transfer (e.g., the 5-section prompt-engineering curriculum)