jgs-v1-diagrams

name	jgs-v1-diagrams
description	Inventory and export SysML v1 diagrams — build a visual review pack of diagram images, list diagram kinds, and compare layout styles. FREE tier (read-only). Requires jgs-magic-sysmlv1-mcp. Trigger: "export diagrams", "diagram images", "PNG", "review pack", "visual review", "diagram inventory", "layout styles", "compare layouts".

jgs-v1-diagrams — Diagram Inventory & Visual Export

You are a context-free AI agent executing the jgs-v1-diagrams skill. Read-only — never call enable_writes or any mutation tool. Any image files are saved with the agent's own Write tool.

Invocation: /jgs-v1-diagrams [<package-name-or-id>]

Behavior-content caveat (bridge gap, 2026-06-17): list_children / get_element_structure / walk_tree do NOT enumerate Activity.node/edge or StateMachine region/state/transition collections — a populated activity/state-machine reports childCount:1. Never judge an Activity or StateMachine "empty" from those tools. Judge behavior from the exported PNG, get_model_metrics type counts (OpaqueActionImpl/ControlFlowImpl/TransitionImpl/StateImpl), or execute_groovy getNode()/getRegion().getTransition().

---

Tool constraints — read before calling anything

• list_diagrams requires parent_id — there is no model-wide form. Call it once per package node.
• walk_tree takes root_id/max_depth/max_elements; pass max_elements: 2000 and warn if the result count equals 2000 (possible truncation).
• export_diagram_image takes a diagram_id and returns an inline base64 JPEG (max 1024px) as MCP ImageContent — it does NOT return a PNG and does NOT write a file. To persist it, decode the base64 and write a .jpg with the agent's Write tool.
• list_diagram_symbols, list_layout_styles, compare_layout_styles all take a diagram_id.
• list_diagram_kinds takes no arguments (reference list of supported kinds).

---

Step 1 — Resolve scope

• No argument: call mcp__jgs-sysmlv1__get_root_package → root_id.
• Argument given: resolve via find_by_name/find_by_qualified_name (handle 0 / 1 / many / 50-cap as in jgs-v1-navigate) → root_id.

Step 2 — Collect package nodes

Call mcp__jgs-sysmlv1__walk_tree({"root_id": "<root_id>", "max_depth": 10, "max_elements": 2000}). Collect every package node ID. (Truncation warning if count == 2000.)

Step 3 — Inventory diagrams

For each package node, call mcp__jgs-sysmlv1__list_diagrams({"parent_id": "<package_id>"}). Collect diagram id, name, kind, owning package. For symbol counts, call mcp__jgs-sysmlv1__list_diagram_symbols({"diagram_id": "<diagram_id>"}) per diagram. Caveat: list_diagram_symbols is a partial ("Phase B") capability and may return 0 symbols for a diagram that actually has content (confirmed live). So do NOT definitively label a diagram a stub on a low/zero count — render the symbol count as n/a when 0 and treat "stub" as a soft hint only.

Optionally call mcp__jgs-sysmlv1__list_diagram_kinds once to annotate/validate the kinds seen.

Step 4 — Export images (visual review pack)

Ask the user: "Export images for all N diagrams, or a subset? (all / list names)". For each selected diagram, call mcp__jgs-sysmlv1__export_diagram_image({"diagram_id": "<diagram_id>"}). Decode the returned base64 JPEG and write it as jgs-v1-diagrams/<sanitised-diagram-name>.jpg with the agent's Write tool. If an export fails, note it and continue.

Step 4b — Visual self-check (vision; bounded)

Exporting proves the diagram rendered — not that it is readable. After Step 4, if you can read images, score each exported JPEG against the shared defect taxonomy and report a verdict. Full taxonomy + round discipline: _magic-sysmlv2-self-verify/reference/visual-self-check.md (v1 flavour).

Quick taxonomy: symbol overlap · edge spaghetti · label truncation · cramped density · wrong orientation/aspect · disconnected/gridded boxes (no drawn paths) · blank/near-blank (→ AMBIGUOUS, not a defect) · low contrast.

• Read-only skill: you may NOT re-layout (no auto_layout_diagram — that's a write). So here the self-check is diagnostic only: report defects per diagram in the output pack ("⚠️ overlap", "⚠️ labels clipped", or "clean"); recommend the user re-run layout in a PRO/write context. Max 1 scoring pass (no fix loop — you can't fix read-only).
• Blank/stub image → AMBIGUOUS ("diagram may not be open / export failed"), never a clean verdict and never an "empty diagram" claim.
• No vision → skip; show the images and state the visual layer was not auto-checked.

Add a Visual column to the output table (clean / ⚠️ / n/a).

Step 5 — (optional) Layout comparison

If the user asks about layout, call mcp__jgs-sysmlv1__list_layout_styles({"diagram_id": "<id>"}) and mcp__jgs-sysmlv1__compare_layout_styles({"diagram_id": "<id>"}) for a chosen diagram and summarise.

---

Output Format

## Diagram Review Pack — <model or package name>

**Diagrams:** <total> (<count> by kind…)
**Exported:** <n> images → ./jgs-v1-diagrams/
**Stub diagrams (<3 symbols):** <list or "none">

| Diagram | Kind | Package | Symbols | Image | Visual |
|---|---|---|---|---|---|
| <name> | <kind> | <pkg> | <n> | <file.jpg or — not exported —> | clean / ⚠️ <defects> / n/a |

---

Error handling

• Bridge not reachable: "jgs-magic-sysmlv1-mcp bridge is not reachable. Is CATIA Magic running with the bridge plugin active?" Stop.
• list_diagrams fails for a package: note inline and continue with other packages.
• export_diagram_image fails for a diagram: mark it "— export failed —" and continue.

What not to do

• Do not call list_diagrams without parent_id.
• Do not describe the export output as PNG — it is JPEG (≤1024px).
• Do not call any model write tool.

Common Mistakes

Mistake	Fix
list_diagram_symbols returns 0 → diagram called a "stub"	0 is a known partial-capability result, NOT proof of emptiness — render the count n/a, treat "stub" as a soft hint only
Activity / StateMachine reported "empty" (childCount:1)	list_children/get_element_structure/walk_tree do NOT enumerate Activity nodes/edges or SM regions/transitions — judge from the exported image, get_model_metrics type counts, or execute_groovy getNode()/getRegion().getTransition()
Treating export_diagram_image output as a saved file	It returns an inline base64 JPEG (≤1024px) — decode and Write it yourself to persist
list_diagrams called model-wide	It requires parent_id — call once per package node
Calling a blank export an "empty diagram"	Confirm it's OPEN; a closed/failed export is AMBIGUOUS, not an empty diagram