id: WO-0001
title: Enterprise Architecture Agent for STM
requestor: CEO
assignee: Clara
participants: []
status: completed
created: 2026-03-18
completed: 2026-03-20
impacts:
process: production-pipeline
capability: enterprise-architecture
roles: [Pablo, Clara, Quinn]
Work Order: Enterprise Architecture Agent for STM
Order ID: WO-0001
Client: STM (Societe de transport de Montreal)
Product type: Enterprise Architecture Agent
Production line: digital-talent
Status: Completed
Date: 2026-03-18
1. Client Profile
| Field |
Value |
| Client |
STM (Societe de transport de Montreal) |
| Segment |
Public transit authority, 50-500 employees |
| Need |
EA guidance without a full-time enterprise architect |
| EA maturity |
Existing practice with Macroscope STM + TOGAF ADM |
| IT landscape |
Heterogeneous, medium-high complexity |
| Repository |
SAP LeanIX v4 |
| Language |
French Canadian |
| Documentation |
Confluence |
| Business process |
Architecture governance -- request intake, orientation, solution design, validation, publication |
| Business capability |
Enterprise Architecture practice management and delivery |
| Key user roles |
Enterprise Architect (primary), Solution Architects, Technical Architects, IT Directors (consumers) |
| AI platform |
Claude Code (Claude CLI agent with skills/commands) |
2. Product Definition
The EA digital talent will provide 9 capabilities:
| # |
Capability |
Description |
| 1 |
Requirements capture |
Structured intake from voice notes, meeting transcripts, or written briefs into formal EA intrants |
| 2 |
Architecture orientation |
Multi-perspective analysis (EA/SA/TA trio) producing orientation documents with recommendations |
| 3 |
Solution design |
Architecture solution documents with impact analysis, integration patterns, publication-ready output |
| 4 |
Diagram generation |
TOGAF-aligned diagrams (L1-L7) in Draw.io format from structured inputs |
| 5 |
Architecture decision records |
ADR capture with options analysis, rationale, and consequences |
| 6 |
Quality validation |
Repository Fact Sheet validation, CRA dossiers, architecture compliance checks |
| 7 |
Cost-benefit analysis |
ROI calculations, breakeven analysis, investment recommendations |
| 8 |
Roadmap planning |
Delivery strategy, phases, milestones, Gantt-style roadmaps |
| 9 |
Publication |
Documentation-platform-ready output with diagram macro integration |
3. Deliverable Set
| Artifact |
Description |
| CLAUDE.md |
Client-specific agent configuration: role definition, workspace rules, conventions, model selection |
| .claude/commands/*.md |
12-15 EA skill commands tailored to Macroscope STM + TOGAF |
| .claude/commands/templates/ |
Draw.io diagram templates (9 types) |
| content-in/methodology/ |
Macroscope STM methodology summary, terminology, process lifecycle |
| content-in/meta-model/ |
LeanIX v4 data model, validation rules, field definitions |
| content-in/tools/ |
Verification checklists, quality criteria, publication formatting |
| content-in/requests/ |
Request template, per-request CLAUDE override template |
| workflows/ |
HLSD orchestrator workflow |
| demandes-ae/ |
Per-request working directory structure |
| Client documentation |
User guide, configuration guide, skill reference card |
4. Skill Inventory
| # |
Skill |
File |
Category |
Description |
| 1 |
Note review |
ea-exigences-note-revue.md |
Requirements |
Structured Q&A from meeting notes |
| 2 |
Note processing |
ea-exigences-note.md |
Requirements |
Raw note capture and formatting |
| 3 |
Intrant generation |
ea-exigences-intrant.md |
Requirements |
Complete EA intrant from reviewed notes |
| 4 |
Architecture orientation |
ea-archi-orientation.md |
Analysis |
Multi-perspective analysis with recommendation |
| 5 |
Architecture catalog |
ea-archi-catalogue.md |
Analysis |
Repository catalog updates |
| 6 |
Solution design |
ea-changement-solution.md |
Solution |
Architecture solution with impact analysis |
| 7 |
Cost-benefit analysis |
ea-changement-couts.md |
Solution |
ROI and breakeven analysis |
| 8 |
Roadmap |
ea-changement-feuilleroute.md |
Solution |
Delivery roadmap with milestones |
| 9 |
Repository validation |
ea-qualite-validation.md |
Quality |
LeanIX Fact Sheet completeness checks |
| 10 |
CRA dossier |
ea-qualite-cra.md |
Quality |
Architecture compliance dossier |
| 11 |
Diagram generation |
ea-commun-diagramme.md |
Common |
Draw.io diagram from structured input |
| 12 |
Publication |
ea-commun-publication.md |
Common |
Confluence-ready output formatting |
| 13 |
Decision record |
kdd.md |
Management |
ADR with options, rationale, consequences |
| 14 |
Planning |
plan.md |
Management |
Structured work plan for complex requests |
| 15 |
Orchestrator |
ea-hlsd.md |
Orchestration |
End-to-end workflow orchestration |
5. Pattern Selection
Primary Patterns (6)
| Pattern |
Category |
Why |
| Prompt Chaining |
Orchestration |
Sequential pipeline: note -> Q&A review -> intrant -> validate -> diagram -> solution. Each output feeds the next. |
| Orchestrator-Workers |
Orchestration |
HLSD orchestrator coordinates 12+ specialist skills, detects state, plans remaining work. |
| Routing |
Orchestration |
Agent analyzes incoming requests (voice note? orientation question? solution request?) and routes to the correct skill. |
| Tool Use |
Tool Use |
Heavy file I/O (reading intrants, writing deliverables), Draw.io generation, CSV parsing, publication formatting. |
| Quality Gates |
Orchestration |
Mandatory validation checkpoint between intake and solution phases -- workflow halts if quality is insufficient. |
| Plan-Then-Execute |
Reasoning |
/plan skill creates structured work plan before execution. Orchestrator detects state and plans remaining steps. |
Supporting Patterns (6)
| Pattern |
Category |
Why |
| Evaluator-Optimizer Loop |
Reasoning |
Repository validation iterates: identify gap -> suggest fix -> re-validate. |
| Guardrails and Safety |
Tool Use |
Methodology compliance, notation correctness, safe file operations. |
| Memory Management |
Memory |
Per-request folder structure (demandes-ae/) persists context across sessions. |
| Escalation |
Human-in-the-Loop |
Escalates when 3+ viable architecture options with significant tradeoffs. |
| Approval Gate |
Human-in-the-Loop |
Human must review validation results before diagram/solution generation. |
| Feedback Loop |
Human-in-the-Loop |
Post-delivery feedback collection and routing. |
Patterns NOT Used (and why)
| Pattern |
Why Not |
| Parallelization |
EA work is inherently sequential -- each artifact depends on the previous. |
| Multi-Agent Coordination |
Single-agent product. The EA talent works alone in the client environment. |
| Shared State |
Single agent, no shared state needed. File system serves as persistence. |
| RAG |
Not needed for v1. Context window handles reference materials. Future enhancement for large knowledge bases. |
| Reflection |
Covered by the Evaluator-Optimizer pattern via repository validation. |
| Fallback and Recovery |
Low risk of tool failures in file-based workflow. Standard error handling sufficient for v1. |
6. Customization Values
| Dimension |
STM Value |
| Methodology |
Macroscope STM + TOGAF ADM |
| Language |
French Canadian |
| Repository tool |
SAP LeanIX v4 |
| Diagram standard |
ArchiMate 3.x via Draw.io |
| Publication target |
Confluence |
| Deliverable codes |
A100, A230, A270, A280, A290, A620 |
| Request structure |
DAE-NNNN |
| Decision format |
KDD-NNN |
| Model selection |
Haiku for execution, Sonnet for analysis, Opus for orchestration |
| Quality thresholds |
LeanIX Fact Sheet completeness |
| Skill set |
Full (15 skills) |
7. EA-Specific Intake Questions
Supplement the generic intake questions (Stage 1) with these EA-specific questions:
Methodology and Frameworks
- Which EA methodology do you follow? (TOGAF, Zachman, Macroscope, custom, none)
- Do you have an existing EA repository tool? (LeanIX, Ardoq, Mega, none)
- What diagramming standard do you use? (ArchiMate, UML, informal, flexible)
- Do you follow a specific notation for architecture diagrams?
Current State
- Do you have existing architecture documentation? What format?
- How many applications/systems are in your landscape? (order of magnitude)
- What are your top 3 architecture pain points today?
- How are architecture decisions currently made and recorded?
Desired Capabilities
- Which EA deliverables do you need most? (architecture reviews, integration guidance, technology evaluation, decision records, roadmaps)
- How do you want to receive outputs? (Confluence, SharePoint, Notion, markdown files)
- Do you need diagram generation? Which types?
- Do you need cost-benefit / ROI analysis capabilities?
Operational
- What language should the agent work in?
- How many architects or technical staff will use the agent?
- What is your documentation platform?
- Do you have naming conventions for artifacts, requests, and decisions?
Scope Boundaries
- What should the agent NOT do?
- Are there compliance or regulatory constraints on architecture decisions?
- What is your timeline expectation?
- What does success look like at 30 days? 90 days?
8. EA-Specific QA Test Cases
Functional Tests (8 cases, must pass all)
| # |
Test Case |
Pass Criteria |
| F1 |
Note review |
Produces structured Q&A from sample meeting notes |
| F2 |
Intrant generation |
Produces complete EA intrant with all sections populated |
| F3 |
Architecture orientation |
Multi-perspective analysis with justified recommendation |
| F4 |
Solution design |
Architecture solution with impact analysis, publication-ready |
| F5 |
Diagram generation |
Valid Draw.io XML with correct ArchiMate notation |
| F6 |
Decision record |
ADR with options, rationale, consequences |
| F7 |
Repository validation |
Detects known gaps in test repository data |
| F8 |
Orchestrator |
Correct state detection, proper sequencing, quality gate halts |
Edge Case Tests (6 cases, must pass 5/6)
| # |
Test Case |
Expected Behavior |
| E1 |
Empty input |
Reports missing input, no garbage output |
| E2 |
Ambiguous request |
Asks clarifying questions or escalates |
| E3 |
Conflicting requirements |
Identifies conflict, flags for human decision |
| E4 |
Large input (10+ pages) |
Handles without truncation |
| E5 |
Wrong skill invoked |
Redirects to correct skill |
| E6 |
Incomplete orchestrator state |
Detects gap and halts clearly |
Documentation Tests (3 cases, must pass all)
| # |
Test Case |
Method |
| D1 |
User guide accuracy |
Follow workflows, verify behavior matches |
| D2 |
Skill reference accuracy |
Compare descriptions to actual I/O |
| D3 |
CLAUDE.md accuracy |
Verify skills table matches .claude/commands/ |
Scoring
PASS: Functional 8/8 AND edge case >= 5/6 AND documentation 3/3
CONDITIONAL: Functional 8/8 AND edge case 4/6 with remediation plan
FAIL: Any functional failure OR edge case < 4/6
9. Build Estimate
| Component |
Effort |
| Phase A: Repository setup |
30 min |
| Phase B: CLAUDE.md + core config |
2-4 hours |
| Phase C: Skills (15 x 0.5-1 hr) |
5-14 hours |
| Phase D: Templates (9 Draw.io + 3 doc) |
3.5-10 hours |
| Phase E: Reference materials |
1-2 hours |
| Phase F: Orchestrator + workflow |
2-3 hours |
| Phase G: Documentation |
2-3 hours |
| Phase H: Verification |
1-2 hours |
| Total |
14-26 hours (2-4 days) |
10. Reference Implementation
STM production instance: C:/Projects/agent-ea/
This is the existing EA agent built for STM before the factory existed. It serves as the reference to clone and adapt. The first production run through the line should start from this reference, validate it against the production line quality standards, and package it properly.
11. Dependencies
| Dependency |
Source |
Status |
| Pattern catalog |
Ada (Agentic Pattern Designer) |
Available -- 15 patterns |
| Production line blueprint |
Elena (Enterprise Architect) |
Available |
| QA criteria |
Quinn (QA Engineer) |
Needs EA-specific test data |
| Deployment checklist |
Diego (Deployment Specialist) |
Needs EA-specific checks |
| Client intake process |
Camille (Client Intake Manager) |
Needs EA-specific questions (Section 7 above) |
| Reference implementation |
agent-ea (STM) |
Available at C:/Projects/agent-ea/ |
12. EA-Specific Feedback Questions
At each feedback milestone, supplement generic questions with:
- Requirements Capture: Frequency of use, output quality (1-5), missing fields?
- Architecture Analysis: Orientation doc usefulness (1-5), viewpoint coverage, methodology gaps?
- Solution Design: Format effectiveness for stakeholders, impact analysis depth, publication issues?
- Diagrams: Most-used types, Draw.io quality, missing diagram types?
- Decision Records: Usage frequency, options analysis format effectiveness?
- Quality Validation: Catching real issues? False positives? Missed gaps?
- Overall: Satisfaction (1-10), top change request, NPS recommendation score
13. Template Library
Draw.io Diagram Templates
| Template |
TOGAF Level |
Purpose |
| drawio-business-context.drawio |
L1 |
High-level business context |
| drawio-architecture-applicative.drawio |
L3 |
Application landscape |
| drawio-integration-card.drawio |
L4 |
System integration detail |
| drawio-er-model.drawio |
L5 |
Entity-relationship/data flow |
| drawio-bpmn-process.drawio |
L6 |
Business process flow |
| drawio-interaction-application.drawio |
L3 |
Application interactions |
| drawio-roadmap-swimlane.drawio |
-- |
Delivery roadmap (org lanes) |
| drawio-roadmap-timeline.drawio |
-- |
Timeline-based plan |
| gabarit-business-card-3layers.drawio |
L1 |
Layered capability card |
Document Templates
- Request template (demandes-ae/ folder structure)
- CLAUDE-per-request override template
- Verification checklist