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.
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:
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).
The request updates an existing file → route by its extension: .drawio → drawio-generator,
.excalidraw → excalidraw-generator.
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.
The request names both tools → if it asks for both files, follow Edge Cases → Both formats;
otherwise ask the routing question in rule 6.
The request asks for a hand-drawn, sketch, whiteboard, or wireframe look → excalidraw-generator.
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
With dg_mode=lease, run asm deps acquire <engine> --session <dg_session> --json for the
routed engine only. Record the returned skillMdPath.
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.
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.
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
Apply the routing rules. If rule 1 matches, write the out-of-scope Final Report and stop.
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.
Print one line: Routing to <engine>: <the rule that matched, in the user's words>.
Run the Dependency Preflight.
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.
Relay the engine's Final Report as the run's closing output (Final Report below).
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 --session <dg_session> --json.
When the umbrella stops before any engine runs (out of scope, unanswered routing question, missing
engine), it writes its own four-line report with status BLOCKED. Templates and fill rules:
references/final-report.md. Expected output when the routed engine is missing:
Result: BLOCKED. No diagram written: drawio-generator is not installed.
Evidence: Routed to drawio-generator (rule 3, "in draw.io"). Preflight dg_mode=lease; asm deps acquire drawio-generator failed.
Uncertainty: No engine ran, so no file was generated or validated.
Decision: Install drawio-generator with the printed command, then re-run. Excalidraw output is available only if you ask for it.
Example
Input: "Draw a sketchy onboarding wireframe for mobile."
Route: Routing to excalidraw-generator: rule 5, "sketchy wireframe".
Closing: excalidraw-generator's Final Report, relayed unchanged, e.g.
Result: COMPLETE. Wrote mobile-onboarding-wireframe.excalidraw (...)
Acceptance Criteria
Verify these for every run:
Routing follows the first matching rule. Exactly one engine runs, unless the user asked for
both files.
The routed engine matches the requested tool, existing file extension, or aesthetic, and the
Routing to line names the matching rule.
The preflight acquires or checks only the routed engine. A missing engine stops the run with
no file written and no substitution.
When an engine ran, the closing output is that engine's Final Report, unchanged.
When the umbrella stopped, its report opens with Result: BLOCKED and has Evidence:,
Uncertainty: and Decision: lines.
The closing report passes the four reader checks in references/final-report.md (result
findable, facts and assumptions separated, claims traceable, next decision clear). Without a
human reviewer's answer, human understanding is unconfirmed.
Edge Cases
Both formats — run the engine the user named first, or drawio-generator when no order was
given. After its Final Report, run the other engine on the same content without asking again,
because the user already asked for both: acquire it under the same dg_session. If the first
engine ended BLOCKED, ask before running the second. Relay each engine's Final Report in turn.
Unanswered routing question — ask it once more. Silence or a timeout is not delegation. If the
user still does not answer, write the unanswered-routing Final Report.
Routed engine unavailable — the preflight stops the run. Name the other engine as an option
the user may request explicitly, and never switch to it on your own.
Request for an image file (PNG, SVG) — say that both engines write an editable file that its
editor can export to PNG or SVG. If the user accepts that, continue with the routing rules.
Otherwise write the out-of-scope Final Report.
Create, update, or audit AGENTS.md (default) or CLAUDE.md agent instructions: set up, prune, add a rule, or merge CLAUDE.md or .cursorrules into one file Claude Code, Codex, and Cursor all read. Don't use for READMEs, skills, or subagents.
Find open AI engineering jobs that fit a candidate, verify location and apply links, and assess each employer's product. Use for one-off shortlists or a job alert's research step. Don't use for résumé writing, job applications, or company news.
Generate App Store screenshots and header/search images for iPhone, iPad and Mac from a codebase or landing page, or re-render a set. Use when preparing a submission. Don't use for ASO keywords, review audits, icons, videos, or Google Play.
Audit iOS/macOS apps against App Store Review Guidelines before submission, with evidence-backed verdicts and fixes. Don't use for Google Play, general code review, or rejection appeals.
Optimize App Store and Google Play listings via a 7-phase, plan-approval-gated ASO workflow (keyword strategy, metadata, localization). Don't use for web SEO, paid UA campaigns, or pre-submission compliance audits.
Generate a commit message, stage all changes, and push to remote after scanning for secrets, large files, and protected-branch risks. Skip for opening PRs, code review, or cutting releases/tags.