Skip to content

mattgierhart/PRD-driven-context-engineering / temp/v0.5-requirements/prd-v05-technical-stack-selection

prd-v05-technical-stack-selection

Verified

Make technology decisions for every product capability by discovering existing assets, evaluating vendor-aligned options, and categorizing as Reuse/Extend/Build/Buy/Integrate/Research during PRD v0.5 Red Team Review. Handles both greenfield and brownfield contexts. Triggers on "tech stack", "build or buy?", "what technologies?", "technical decisions", "what do we reuse?", "existing stack", "vendor constraint", "IBM-first", "what tools do we need?", "evaluate solutions", "select tech stack". Consumes FEA- (features), SCR- (screens), RISK- (constraints). Outputs TECH- entries with decisions, rationale, and cross-references. Feeds v0.6 Architecture Design.

FileKindSize
LEARNINGS-LIGHTSTACK.mdmarkdown8.7 KB
SKILL.mdskill md14.0 KB
assets/evaluation-scorecard.mdmarkdown2.0 KB
assets/tech-reuse.mdmarkdown1.7 KB
assets/tech.mdmarkdown2.2 KB
references/brownfield.mdmarkdown3.6 KB
references/examples.mdmarkdown5.5 KB
references/ibm-products.mdmarkdown4.7 KB

references/examples.md

View on GitHub
# TECH- Entry Examples

Three filled examples covering the most common decision categories: Reuse, Buy, and Research.

---

## Example 1: Reuse Decision

```
TECH-109: Application Authentication & SSO
Category: Reuse
Layer: Auth
Purpose: Enterprise SSO, user identity, role-based access control

Reuse From: enRoll — Keycloak deployment on IKS
Reuse Scope: Shared instance (separate realm for INTERPRET)
Current State: Keycloak is deployed and operational for enRoll on IBM Cloud
  Kubernetes. Team has existing operational experience. Supports SAML 2.0,
  OIDC, LDAP/AD federation.
Changes Needed: Create dedicated INTERPRET realm. Configure health plan
  IdP federations for INTERPRET customers. Add INTERPRET-specific roles
  (analyst, reviewer, admin).

Shared or Separate: Shared deployment, separate realm

Features Served: FEAT-103 (RBAC for approve/reject), all user-facing features
Risk Constraints: Enterprise customers require SSO

Cost: $0 (already running on IKS)
Integration Complexity: Low (team already operates Keycloak)
Lock-in Risk: Very Low (open-source, self-hosted, Red Hat project)

Product Family Notes: Shared Keycloak enables single sign-on across
  LightStack products if a customer uses both enRoll and INTERPRET.
  Consider if customers should have unified identity across products.
```

**Why this is a good Reuse entry:**
- Documents where the asset comes from and what state it's in
- Explicitly lists what changes are needed (not just "reuse it")
- Addresses shared vs. separate deployment
- Explains the product family implication (cross-product SSO)
- Cost is $0 because it's already running

---

## Example 2: Buy Decision

```
TECH-105: Primary Database
Category: Buy
Layer: Database
Purpose: Rule storage, document metadata, user data, version history, audit trails

Features Served: FEAT-100, FEAT-101, FEAT-102, FEAT-103, FEAT-104,
  FEAT-200, FEAT-201
Risk Constraints: RISK-006 (HIPAA onshore processing)

Decision: IBM Cloud Databases for PostgreSQL
Rationale: IBM-first constraint satisfied. HIPAA-ready with BAA. Dallas TX
  region meets onshore requirement. Standard PostgreSQL — no proprietary
  extensions. Supports pgvector extension for vector search (see TECH-106).
  JSONB handles semi-structured rule and document metadata.

Alternatives Considered:
  - Supabase: Higher developer velocity (auto-generated APIs, auth, real-time)
    but non-IBM, adds external dependency for HIPAA data
  - Self-managed PostgreSQL on IKS: Maximum control, cheapest long-term,
    but significant ops burden for a small team
  - MongoDB on IBM Cloud: Document store fits policy data shape but weaker
    for relational queries (rule → source linking)

Trade-offs:
  - Pro: IBM-first, HIPAA-ready, standard PostgreSQL, supports pgvector
  - Con: Lower developer velocity than Supabase, enterprise pricing

Cost: ~$100-300/month current scale. ~$500-1,500/month at 10x.
Integration Complexity: Low
Lock-in Risk: Low (standard PostgreSQL — portable to any PostgreSQL host)
```

**Why this is a good Buy entry:**
- Rationale explains WHY, not just WHAT (IBM-first + HIPAA + pgvector)
- Alternatives include a genuinely better option (Supabase) with honest rejection reason
- Cost includes both current AND 10x scale
- Lock-in risk is assessed with mitigation (standard PostgreSQL = portable)

---

## Example 3: Research Decision

```
TECH-100: AI Foundation Platform
Category: Buy (constrained) + Research
Layer: AI/ML
Purpose: Foundation model access, fine-tuning, RAG for policy interpretation

Features Served: FEAT-101 (AI Policy Interpretation), FEAT-102 (Citation
  Linking), FEAT-204 (Continuous Learning)
Risk Constraints: RISK-001 (AI accuracy target), RISK-003 (citation fragility),
  RISK-005 (watsonx.ai platform lock-in)

Decision: watsonx.ai as primary with model-agnostic abstraction layer
Rationale: Business constraint (IBM-first). Model Gateway provides multi-model
  access. RISK-005 mitigation requires abstraction layer to enable benchmarking
  against alternatives without architectural rework.

Alternatives Considered:
  - Azure OpenAI: Strong healthcare benchmarks (HealthBench). Consider for
    POC benchmark comparison.
  - Anthropic Claude (via Bedrock): Strong document analysis capabilities.
    Consider for POC benchmark comparison.
  - Direct OpenAI: GPT-4o strong on extraction tasks. Not IBM-aligned.

Trade-offs:
  - Pro: IBM relationship, Granite models, integrated governance ecosystem
  - Con: Healthcare policy accuracy unproven (RISK-001), lock-in without
    abstraction layer (RISK-005)

Cost: Enterprise pricing (existing IBM relationship)
Integration Complexity: Medium (SDK integration + abstraction layer)
Lock-in Risk: High without abstraction layer; Medium with it

Research Needed:
  - Benchmark watsonx.ai vs Azure OpenAI vs Claude on 5-10 real policy documents
  - Measure: rule extraction accuracy, citation quality, cost per document
  - Test fine-tuning ROI on healthcare policy domain

Evaluation Criteria:
  - Rule extraction accuracy >= 85% (go/no-go threshold)
  - Citation accuracy >= 80% at paragraph level
  - Cost per 100-page document < $50

Decision Deadline: After POC (before v0.7 Build Execution)
```

**Why this is a good Research entry:**
- Makes a default decision (watsonx.ai) while acknowledging uncertainty
- Research scope is specific (5-10 real docs, 3 providers, 3 metrics)
- Evaluation criteria are measurable (85% accuracy, $50/doc)
- Decision deadline is tied to a lifecycle milestone (before v0.7)
- Cross-references 3 RISK- entries that drive the research need
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 322kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 197kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 197kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 197kUpdated Claude CodeCodexVerifiedInstall