القائمة
Generate clean architecture / flow diagrams as PNG+SVG for the yai stack. Hand-laid-out Python→SVG→PNG pipeline (no Mermaid auto-layout). Auto-load when asked to create, update, or render an architecture diagram, flow chart, system diagram, or "draw the stack".
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
agent-browser CLI — headless Chromium for AI agents. Snapshot-and-ref workflow, core commands, and yai service interaction patterns. Auto-load when asked to open a browser, navigate a URL, click, fill, screenshot, or interact with any yai web UI (Grafana, n8n, Windmill, Langfuse, MinIO, etc.).
Browserless headless Chromium service on the yai stack — REST endpoints (screenshot, PDF, scrape, function), WebSocket CDP/Puppeteer/Playwright patterns, token auth, sizing. Auto-load on Browserless, headless browser, Puppeteer, Playwright, CDP, screenshot, PDF, scrape-script questions.
Firecrawl web crawling & scraping API on the yai stack — scrape, crawl, extract endpoints, job status, concurrency knobs, proxy config. Auto-load on Firecrawl, web scrape, crawl, extract, markdown, web content questions.
Grafana on the yai stack — datasource UIDs, file provisioning (not TF), dashboard conventions, scrape targets page, Grafana plugin setup. Auto-load on Grafana, dashboard, datasource, DS_*, panel, templating variable, provisioning questions.
How to control every yai web UI — login recipes, navigation patterns, and common actions for Grafana, Langfuse, MinIO, n8n, Windmill, LiteLLM, Qdrant, and Traefik. Auto-load when asked to log in, navigate, click, or perform any action in a yai service UI.
Langfuse on the yai stack — LLM observability (traces, evals, prompt mgmt), REST API, SDK usage, Langfuse–LiteLLM callback wiring. Auto-load on Langfuse, LLM traces, evals, prompt management, observation, session questions.
| name | diagram |
| description | Generate clean architecture / flow diagrams as PNG+SVG for the yai stack. Hand-laid-out Python→SVG→PNG pipeline (no Mermaid auto-layout). Auto-load when asked to create, update, or render an architecture diagram, flow chart, system diagram, or "draw the stack". |
| when_to_use | Load when the user wants to visualize the yai stack architecture, create a flow diagram, update an existing diagram, or render any system or component diagram as a PNG or SVG file. |
Produce publication-quality architecture diagrams with deterministic, hand-controlled layout. Do not use Mermaid for these — its auto-layout crosses edges and looks messy. Instead generate SVG from a small Python script with explicit coordinates, then rasterize.
python3 gen.py # writes the .svg
rsvg-convert -z 2 out.svg -o out.png # rasterize at 2x
rsvg-convert is preinstalled (/opt/homebrew/bin/rsvg-convert). If a
diagram is needed quickly, copy template.py from this skill folder, edit
the GROUPS / NODES / arrow section, and run it.
The canonical, maintained example lives at docs/architecture.py in the
repo root (the yai stack architecture itself). template.py here is a
frozen copy of that generator — use it as the starting point.
rx=14), light pastel tint fill, 2px colored
border, bold dark title + smaller colored subtitle. Soft drop shadow.rx=20, stroke-dasharray="7 6",
stroke-opacity≈0.55) with an UPPERCASE letter-spaced label in the
top-left corner. Leave ~40px between the label and the first node.#8b96a5) for primary flow; dashed colored for
feedback / telemetry loops. Curved Bézier connectors, arrowhead markers.c (border/text), tint (fill),
dark (title) as a triple. See PAL in the template.Save all three next to each other (the repo uses docs/):
architecture.png (raster), architecture.svg (vector), architecture.py
(generator — so the diagram stays editable).
Regenerate after edits:
python3 docs/architecture.py && rsvg-convert -z 2 docs/architecture.svg -o docs/architecture.png
& to & in every text string (titles, labels, subtitles)
or rsvg-convert fails with xmlParseEntityRef.