Xiyu Wang
Community@shinygua
Xiyu Wang (ShinyGua) publishes multi-stage research pipelines for equity analysis and academic writing, plus a ComfyUI-based image generation and editing skill suite.
Agent Skills by Xiyu Wang
Showing 52 vetted skills indexed across 3 GitHub repositories.
ShinyGuacomfyui-checkpoint
Saves and restores ComfyUI workflow snapshots in a SQLite version store indexed by session_id.
ShinyGuaplanning-visual-tasks
Decomposes natural-language visual instructions into typed Plan envelopes for ComfyUI orchestration.
ShinyGuapreserving-faces
Compiles a face-locking ComfyUI prompt JSON that re-renders non-face regions at low denoise.
ShinyGuamanaging-models
Downloads ComfyUI model weights from policy-approved sources with sha256 verification.
ShinyGuacomfyui-execution
Submits ComfyUI workflows over HTTP and WebSocket, then fetches execution results and errors.
ShinyGuaevaluating-instruction-following
Scores whether a generated image satisfies the original natural-language instruction.
ShinyGuagenerating-images
Compiles a Visual Intent Object into a ComfyUI txt2img prompt JSON using SDXL or FLUX templates.
ShinyGuaevaluating-identity-preservation
Scores whether a transformed image preserves subject identity against a reference image.
ShinyGuarepairing-workflows
Classifies ComfyUI execution errors and plans minimal partial reruns of failed workflow subgraphs.
ShinyGuasearching-comfyui-solutions
Searches bundled templates and allowlisted web sources for ComfyUI workflow recipes with conservative ranking.
ShinyGuaselecting-models
Resolves installed ComfyUI models for a requested capability via the model registry.
ShinyGuatransferring-styles
Compiles a Visual Intent Object into a ComfyUI img2img style-transfer prompt JSON.
ShinyGuainpainting-regions
Compiles a Visual Intent Object into a ComfyUI SDXL inpainting prompt JSON.
ShinyGuareplacing-backgrounds
Compiles a Visual Intent Object into a ComfyUI background-replacement prompt JSON using inpainting.
ShinyGuamm-company-resolver
Resolves a ticker into a company profile, peer set, and market-context link.
ShinyGuamm-orchestrator
Execute a 14-stage research pipeline from config resolution to reflect.
ShinyGuamm-pdf-exporter
Generates JPM-style equity research PDFs from formatted inputs using LaTeX and xelatex.
ShinyGuamm-discussion-moderator
Synthesize analyst memos and cross-critique files into thesis_map.json and debate_summary.md.
ShinyGuamm-market-analyst
Analyze macro and market environment data to generate investment research memos.
ShinyGuamm-valuation-analyst
Analyze equity valuations and justify price targets in analyst memos.
ShinyGuamm-risk-analyst
Identify downside risks and bear cases for an investment thesis using workspace data.
ShinyGuamm-quant-analyst
Compute technical indicators and relative strength from price data into quant_summary.json.
ShinyGuamm-company-analyst
Writes company fundamentals memos and participates in moderator-guided cross-critique debates.
ShinyGuamm-company-desk
Aggregate news, SEC filings, and catalyst calendars for a given ticker.
Frequently Asked Questions About Xiyu Wang
FAQPage SchemaWhat tasks can I automate using ShinyGua's skills?▼
You can automate end-to-end equity research (ticker resolution, SEC filing aggregation, technical and valuation analysis, BUY/HOLD/SELL decisions, PDF reports), academic paper drafting with citation verification, and ComfyUI image generation, inpainting, style transfer, and background replacement with identity-preservation scoring.
Who are these skills designed for?▼
The mm-* skills target equity analysts and investment researchers producing institutional-grade reports. The ar-* skills serve academic researchers in arts and social sciences. The ComfyUI skills target engineers and creators running local SDXL/FLUX image pipelines who need planning, checkpointing, and error repair.
How do the research pipelines run in practice?▼
Each pipeline starts with an init skill (mm-init or ar-init) that bootstraps a workspace and config. An orchestrator then executes staged skills—data desks, analysts, debaters, reviewers—with progress monitors reading status.json and human approval gates before final report or manuscript export.
What are the prerequisites and dependencies?▼
Equity skills rely on NewsAPI, yfinance, and LaTeX/xelatex for PDF rendering. Academic skills need Markdown sources and reference data for verification. ComfyUI skills require a running ComfyUI server, installed SDXL or FLUX models, and huggingface_hub for gated model downloads behind the allow_model_download setting.
How does the ComfyUI suite handle failures and identity drift?▼
The repairing-workflows skill classifies node_errors or WebSocket execution errors and re-runs only the affected subgraph via partial_execution_targets. Evaluating-instruction-following and evaluating-identity-preservation score outputs against the original instruction and reference image before any step is marked done.