EA Skill Reference
All skills used in the EA pipeline, with usage syntax, inputs, outputs, and pipeline position.
Skills are invoked in Claude Code with a leading /.
Pipeline Position Map
/toolkit:note-create ← capture initial context (raw notes)
↓
/toolkit:note-review ← structure requirements (interactive Q&A)
↓
/ea:leanix-catalog-extract ← EXTRACT: source → Object + Relation + CO CSVs
↓
/toolkit:diagram-generate ← GENERATE: CSV → interactive HTML diagram
↓ (optional)
/toolkit:auto-research ← OPTIMIZE: binary eval loop on generated HTML
↓
/docs-publish ← PUBLISH: HTML → JCT portal or intranet
↓ (for catalog sites)
/jct:add-review-panel ← REVIEW: scaffold client validation panel into portal
/ea:leanix-catalog-extract Extract architecture catalog from any source into standardized CSVs
Usage
/ea:leanix-catalog-extract [source-path] Input
Code directory, document folder, existing CSVs, or nothing (interactive)
Output
{slug}-objects.csv, {slug}-relations.csv, {slug}-co.csv (optional)
Extraction Modes (auto-detected):
Code analysis
.cs, .odx, .btm, .xsd, .yaml, .json, .bicep
Scans codebase for components, processes, interfaces
Document analysis
.md, .html, .docx, .xlsx, .csv
Cross-references entities across documents
Migration project
Legacy + Azure files together
Two-pass: as-is + to-be, mandatory CO generation
Validation
Existing objects + relations CSVs
Check fields, IDs, orphans, relationship types
Interactive capture
No path or "interview"
7-round guided interview (scope → capabilities → apps → integrations)
Key Rules:
- • Schema v3.2 — ID prefix must match Type (APP-001 = Type "Application")
- • Every object gets a Macroscope A-code in Tags (A200, A250, A370…)
- • Initiative sweep is mandatory for all AS-IS extractions (captures in-flight + proposed INI)
- • PRS (Sub-Process) always has Parent_ID pointing to a PRO, not an APP
- • Pure CBR stubs (no business logic) documented in APP Notes only — no PRO/PRS row
Migration project? Auto-detected when both legacy files (.odx, .btm, BizTalk .csproj)
and target files (.yaml, .bicep, Logic Apps) are present, or when "migration" appears in source path.
CO generation is mandatory in this mode.
/toolkit:diagram-generate Generate interactive HTML architecture diagrams from Object + Relationship CSVs
Usage
/toolkit:diagram-generate [csv-directory] Input
{slug}-objects.csv + {slug}-relations.csv
Output
Single self-contained HTML — sidebar nav, layer toggles, clickable nodes, detail panels, zoom/pan
View Modes:
logical
Default — uses Display_Name (org alias). For internal operations.
physical
Uses Name (vendor name). For IT ops, procurement.
conceptual
Uses Group (architectural concept). For board/strategy.
combined
"Name (Display_Name)" — for transition docs, handovers.
Style note: The auto card-grid output has been explicitly rejected for polished deliverables.
For architecture diagrams, default to the hand-composed Azure/Microsoft style
(copy
templates/azure-style-architecture-diagram.html). Use CSV→grid for quick analytical inventories only.
/docs-publish Publish generated documentation to HTML, JCT portal, or intranet
Usage
/docs-publish Driven by
departments/client-services/technical-documentation-engineer-remy/publishing-log.mdTargets
HTML files, JCT Cloudflare Pages, Intranet (Astro)
Remy (Technical Documentation Engineer) owns the publishing pipeline and log.
The log tracks what was published, when, and where. Use /docs-pipeline for the full end-to-end orchestration.
/jct:add-review-panel Scaffold a Client Review Panel into a JCT portal repo
Usage
/jct:add-review-panel What it creates
Interactive review panel for clients to validate EA catalog rows per direction
D1 binding
Reads from customer D1 catalog; writes corrections back via a review workflow
Supports the TFD-0019 validation loop: export per-direction validation sheet → client corrects →
re-import into D1. The review panel is the UX surface for that correction workflow.
See CON-0013 and CON-0014 for the full spec.
/toolkit:note-create & /toolkit:note-review Capture and structure raw architecture context before extraction
/toolkit:note-create
Saves raw dictated thoughts as-is — no reformulation.
Use when coming out of a client interview or discovery session.
/toolkit:note-create [dictated text] /toolkit:note-review
Interactive Q&A session on a raw note. Asks clarifying questions and saves
structured Q&A alongside the original. Run before extraction.
/toolkit:note-review [path-to-note] /toolkit:auto-research Self-improving prompt loop — iterate on generated diagrams or HTML against binary eval criteria
Usage
/toolkit:auto-research [target-file-or-dir] Input
Generated HTML diagram + eval criteria file
Output
Iteratively improved HTML + research log
Based on Karpathy auto-research pattern. Runs multiple iterations against binary pass/fail
criteria until all criteria pass or iteration limit is hit.
Eval criteria template: .claude/commands/ea/auto-research/eval-criteria.md
Skill File Locations
| Skill | File Path |
|---|---|
| /ea:leanix-catalog-extract | .claude/commands/ea/leanix-catalog-extract.md |
| /toolkit:diagram-generate | .claude/commands/toolkit/diagram-generate.md |
| /toolkit:auto-research | .claude/commands/toolkit/auto-research.md |
| /toolkit:note-create | .claude/commands/toolkit/note-create.md |
| /toolkit:note-review | .claude/commands/toolkit/note-review.md |
| /docs-publish | .claude/commands/docs-publish.md |
| /docs-generate | .claude/commands/docs-generate.md |
| /docs-pipeline | .claude/commands/docs-pipeline.md |
| /jct:add-review-panel | .claude/commands/jct/add-review-panel.md |