Implementation Plan: PET Training App

Build the Prompt Engineer Training (PET) app with a modular judge pipeline, gamified learning, and enterprise-ready governance hooks.

Provenance & Traceability

Architectural Decisions (ADRs)

ADR ID Decision Title Impact on This Plan
ADR-018 PET App Architecture Modular judge pipeline and gamified learning model.

Product Requirements (PRDs)

PRD ID Requirement Title Satisfied By (SDS) Acceptance Criteria
PRD-004 Prompt Engineer Training (PET) App SDS-011 FEAT-001..FEAT-007

Software Design Specifications (SDS)

SDS ID Service/Component Bounded Context SEA-DSL Spec File Implementation Status
SDS-011 PET Prompt Judge cognitive-extension N/A Designed

Architecture and Design

Design Principles Applied

Dependency Justification

Expected Filetree

1
2
3

/
├── docs/specs/cognitive-extension/prd/004-pet-app.md
└── docs/specs/cognitive-extension/sds/011-pet-prompt-judge.md

Proposed Cycles

Cycle Branch Wave Files Modified Files Created Specs Implemented
C1A cycle/p025-c1a-specs 1 SDS-011, PRD-004, ADR-018 — All specs
C2A cycle/p025-c2a-domain 2 — libs/pet/domain/** Domain model
C3A cycle/p025-c3a-ports 3 — libs/pet/ports/**, libs/pet/adapters/** Ports + fake
C3B cycle/p025-c3b-llm-adapter 3 — libs/pet/adapters/** LLM adapter
C4A cycle/p025-c4a-api 4 — apps/api/src/modules/pet/** API surface

Task Breakdown

Wave 1: Documentation

Wave 2: Domain Model

Wave 3: Ports & Adapters (Parallel)

Wave 4: API Integration

Validation & Verification

Spec Validation

Implementation Validation

Open Questions

  1. Is PET a standalone shell or integrated into a broader Workbench UI? Workbench /pet
  2. What is the canonical rubric schema for judging and how is it versioned? schemas/pet/rubrics/*.yaml + JSON Schema

Risks & Mitigation

Risk Likelihood Impact Mitigation Strategy
Judge output variability hurts learning outcomes Medium Medium Use deterministic eval modes and rule-based scoring where possible.
Privacy risk from stored prompts Medium High Apply encryption and governance policies; scrub sensitive content.

Rollback Strategy

  1. Disable auto-improve and keep feedback read-only while stabilizing the judge.

Linked Specifications

Type ID/Doc Document
ADR ADR-018 docs/specs/shared/adr/018-pet-app-architecture.md
PRD PRD-004 docs/specs/cognitive-extension/prd/004-pet-app.md
SDS SDS-011 docs/specs/cognitive-extension/sds/011-pet-prompt-judge.md