/diagram-generator

name: diagram-generator
description: "Generate a diagram and route to the right engine — draw.io XML (precise, editable, C4, swimlanes) or Excalidraw JSON (hand-drawn, sketch, wireframes). One entry for flowcharts, architecture, ER, sequence, mind maps. Don't use for Mermaid or slides."
license: MIT
effort: high
dependencies:
  - drawio-generator
  - excalidraw-generator
metadata:
  version: 1.4.0
  author: "Luong NGUYEN <luongnv89@gmail.com>"

Diagram Generator

Single entry point for "make me a diagram". This umbrella picks one engine and hands the request to it. Both engines cover the same diagram types (flowchart, architecture, C4, ER, sequence, mind map) through the same four phases (**Understand → Propose → Generate → Validate**); they differ only in **output format and aesthetic**. The umbrella writes no file itself. Its body stays short to protect the agent's context window; report templates live in `references/`.

Which engine?

| Engine | When the user wants... | Output | |---|---|---| | `drawio-generator` | Precise, professional diagrams to edit later in draw.io / diagrams.net / Confluence; official C4 styling; swimlanes; multi-page | kebab-case `.drawio` file (XML) | | `excalidraw-generator` | A hand-drawn, sketchy, whiteboard feel; wireframes; quick collaborative sketches | kebab-case `.excalidraw` file (JSON) |

Apply the routing rules in order and stop at the first rule that matches:

1. The request asks for **Mermaid**, a slide deck, or brand/marketing graphics → do not route. Write the out-of-scope Final Report (Mermaid is native Markdown; use a presentation or design tool for the others). 2. The request updates an existing file → route by its extension: `.drawio` → `drawio-generator`, `.excalidraw` → `excalidraw-generator`. 3. The request names exactly one tool or editing target ("draw.io", "diagrams.net", "Confluence", "Excalidraw") → route to that engine, even when the aesthetic words point to the other one. 4. The request names both tools → if it asks for both files, follow *Edge Cases → Both formats*; otherwise ask the routing question in rule 6. 5. The request asks for a hand-drawn, sketch, whiteboard, or wireframe look → `excalidraw-generator`. 6. No format signal → ask one question: "Precise and editable (draw.io) or hand-drawn sketch (Excalidraw)?" Wait for the answer. If the user explicitly delegates ("just pick", "choose for me", "use the default"), route architecture, C4, ER, sequence, and swimlane diagrams to `drawio-generator`, and wireframes, brainstorms, and mind maps to `excalidraw-generator`. Anything else that is delegated goes to `drawio-generator`.

Silence is not delegation. Ask content questions (entities, relationships, direction) only inside the engine's Phase 1, never in the umbrella.

Dependency Preflight (mandatory)

This skill invokes the two engines declared in frontmatter `dependencies`, but each run needs only the one that routing selects. Run this once, after routing and before the engine starts:

if command -v asm >/dev/null && asm deps --help >/dev/null 2>&1; then
  asm deps discover diagram-generator --json || echo "discover failed; acquire still runs" >&2
  echo "dg_mode=lease"
else
  echo "asm deps unavailable: npm install -g agent-skill-manager@latest" >&2
  echo "dg_mode=installed"
fi
printf 'dg_session=%s\n' "diagram-generator-$(date +%s)-$$"   # record it; reuse it verbatim

1. With `dg_mode=lease`, run `asm deps acquire <engine> --session <dg_session> --json` for the routed engine only. Record the returned `skillMdPath`. 2. With `dg_mode=installed`, run `test -f "$HOME/.claude/skills/<engine>/SKILL.md" || test -f "$HOME/.agents/skills/<engine>/SKILL.md"` and record the path that exists. 3. If step 1 or 2 fails, print `Missing skill: <engine> — install: asm install github:luongnv89/skills:skills/diagram-generator/<engine> -p claude --yes`, then stop with the missing-engine Final Report. Never substitute the other engine. Never imitate the engine inline. 4. **Release in `finally`.** If any acquire ran, run `asm deps release --session <dg_session> --json` once at every terminal outcome, stops included, after the closing Final Report.

Do not acquire the engine that routing did not select.

Repo Sync Before Edits (mandatory)

The umbrella runs no git command and writes no file. The routed engine runs its own *Repo Sync Before Edits* (scoped to the output file's repository, with its conflict and network-failure rules) before it writes. Do not run a second sync here, and do not skip the engine's sync.

Workflow

1. Apply the routing rules. If rule 1 matches, write the out-of-scope Final Report and stop. 2. If rule 4 or 6 asks the routing question, wait for the answer. If the user has not answered after the question was asked twice (*Edge Cases*), write the unanswered-routing Final Report and stop. 3. Print one line: `Routing to <engine>: <the rule that matched, in the user's words>.` 4. Run the *Dependency Preflight*. 5. Read the recorded `SKILL.md` path and follow that engine's workflow with the user's request unchanged. The engine owns its questions, file naming, Repo Sync, validation, and fix cycles. It also asks for confirmation before it overwrites an existing file; never skip or pre-answer that confirmation. 6. Relay the engine's Final Report as the run's closing output (*Final Report* below). 7. Run the release step of the *Dependency Preflight*.

Final Report

Both engines end every run, stops included, with a four-line Final Report (`Result:` COMPLETE | PARTIAL | BLOCKED, `Evidence:`, `Uncertainty:`, `Decision:`). When an engine ran, its Final Report is this skill's closing output. Relay it verbatim: do not restate it, re-grade it, or change its status. If the release step fails, add one line after the closing report, whichever skill wrote it: `Router note: asm deps release failed for session <dg_session>; run asm deps release