QingFeng Li

Use SAELens to train and analyze sparse autoencoders, which break model activations into a small set of interpretable features. Explore pretrained features, evaluate reconstruction quality, troubleshoot training, and test how individual features influence generated text.

• Exploring features in a pretrained SAE
• Training an SAE for a selected model layer
• Investigating inactive features and reconstruction errors

Review specifications, plans, and tasks together before implementation. Identify missing task coverage, duplicate requirements, unclear wording, and conflicting decisions, then receive a prioritized report with recommendations without changing project files.

• Reviewing specifications and plans before implementation
• Finding requirements with no corresponding tasks
• Checking compliance with project principles

Create or improve Claude skills that package specialized knowledge, repeatable workflows, and tool instructions. The guide covers concise instructions, reusable scripts and resources, template initialization, validation, and packaging into a distributable .skill file.

• Turning recurring work into a reusable Claude skill
• Organizing company knowledge and API references
• Improving an existing skill after practical use

Build production systems for machine learning models, from deployment to monitoring and retraining. The skill also guides LLM integration and retrieval augmented generation, which uses retrieved information to support answers, with scripts and architecture references.

• Deploying machine learning models to production
• Setting up model monitoring and retraining
• Building applications that answer using retrieved information

Find reusable Agent Skills by keyword, category, or tag through prompts.chat. Review descriptions and included files, retrieve a chosen skill, and install its instructions, reference documents, scripts, and configuration files for Claude.

• Finding a code review skill
• Searching for testing skills
• Retrieving all files for a specific skill

Clarify an existing feature specification before implementation planning. The skill prioritizes consequential gaps, asks focused questions one at a time, updates the specification with accepted answers, and summarizes unresolved decisions and recommended next steps.

• Resolving unclear requirements before implementation planning
• Defining error behavior and measurable completion criteria
• Recording agreed answers directly in a specification

Analyze biological data in Python, from DNA, RNA, and protein sequences to evolutionary trees and microbial communities. Read and convert files, compare diversity within and between samples, and test differences between groups.

• Processing DNA, RNA, and protein sequences
• Building and comparing evolutionary trees
• Comparing microbial community diversity

Manage Zotero reference libraries through Python: search, create, update, and delete records, organize collections and tags, upload PDF attachments, and export citation data for research workflows.

• Searching and organizing a reference library
• Adding references and updating bibliographic details
• Uploading PDF attachments to references

Analyze git history to map contributors to security-sensitive files, identify reliance on too few people, and find stale areas. Export CSV and JSON results for ownership reviews, graph exploration, and comparison with CODEOWNERS.

• Find stale security-sensitive code with few maintainers
• Check reliance on individual authentication or cryptography maintainers
• Compare CODEOWNERS with actual contribution history

Inspect production errors in Sentry through read-only API queries. Summarize unresolved issues, occurrence counts, and recent events, then examine individual issue and event details to understand what has been happening.

• Reviewing unresolved production issues from the last 24 hours
• Checking occurrence counts and when errors were last seen
• Inspecting events linked to a specific issue

Analyze how long events take to occur, including records where observation ends before the event. Use Python to prepare data, fit and compare survival models, and evaluate predictions with measures suited to incomplete observations.

• Analyze time to events with incomplete observations
• Compare Cox models and Random Survival Forests
• Select features in datasets with many variables

Configure machine learning training and batch jobs across multiple clouds with SkyPilot. The skill guides GPU selection, lower-cost cloud and region choices, and automatic recovery when discounted spot instances are interrupted.

• Choose a lower-cost cloud for GPU training
• Run long training jobs with recovery after interruptions
• Configure distributed training across multiple machines

Capture desktop screenshots on macOS, Linux, and Windows. Save the full screen, a selected window, or a region defined by coordinates, with support for a specified file path, the default location, or temporary storage.

• Saving a screenshot of the entire desktop
• Capturing a specific app or window
• Saving only a region defined by coordinates

Explore research ideas through conversation, question existing assumptions, and connect methods across disciplines. The skill helps identify research gaps, compare ways to combine papers, and summarize promising directions with practical next steps.

• Explore new research topics and directions
• Find connections across scientific disciplines
• Question assumptions in existing research

Evaluate scientific papers and claims by examining study design, statistics, bias, and the strength of supporting evidence. Helps distinguish supported conclusions from overstatements and provides constructive recommendations for improving research methods and interpretation.

• Assessing the evidence behind a paper's conclusions
• Finding bias and statistical interpretation errors
• Comparing studies with conflicting findings

Create scientific diagrams from written descriptions using Nano Banana Pro. Gemini 3 Pro reviews accuracy, labels, and layout, triggering regeneration when scores fall below the chosen document threshold, up to the iteration limit.

• Creating neural network architecture diagrams for papers
• Illustrating biological signaling pathways
• Preparing study participant flowcharts

Design and improve prompts and structured responses for AI applications. The skill also guides evaluation, retrieval augmented generation, which uses retrieved material to inform answers, and agent architecture, with references and scripts for production workflows.

• Improving prompts and response formats
• Evaluating answers based on retrieved documents
• Designing AI agent systems

Build Python simulations with SimPy to study queues and shared resources such as machines, staff, or servers. Model events over time, measure waiting times and utilization, and compare capacity choices.

• Compare service waiting times under different staffing levels
• Measure machine utilization on a production line
• Model inventory flows between producers and consumers

Plan and audit preprocessing for grouped time-series data, such as matches or patient records. Check missing-value handling, scaling scope, feature meaning, and data leakage, where information unavailable at prediction time enters model inputs.

• Planning missing-value handling within each group
• Choosing within-group or global scaling
• Auditing preprocessing code for data leakage

Create SEO and GEO reports about visibility in search engines and AI answers. Organize findings for executives, technical teams, or clients, with charts, interactive HTML elements, and outputs including Markdown, PDF, JSON, and Excel.

• Prepare an executive summary of SEO and AI search performance
• Compare AI citation trends and competitor performance
• Share an HTML report with charts and data filters

Prepare a complete scholarly publication package, from manuscript sources and submission attachments to reviewer responses and final files. Organize journal or conference requirements, figure specifications, revision tracking, and repeatable build steps around one authoritative manuscript source.

• Preparing a journal or conference submission package
• Organizing point-by-point responses to reviewers
• Checking final files before publication

Create slides for research talks, conferences, and thesis defenses. Plan the narrative and timing, incorporate figures and citations, and produce PDF slides or editable PowerPoint presentations with consistent visual design.

• Prepare a conference talk around key research findings
• Build a thesis defense presentation
• Adapt research figures for a seminar

Build image and video AI systems for object detection and segmentation, which identifies separate image regions. The skill covers custom model training, dataset pipelines, inference optimization, real-time processing, and deployment and monitoring in production.

• Developing object detection and image segmentation systems
• Training custom vision models with project data
• Optimizing inference for real-time video processing

Create focused checklists to review requirements before implementation. Using project specifications and plans, the skill turns gaps, vague wording, conflicting requirements, and missing scenarios into questions that help authors and reviewers assess specification quality.

• Finding missing requirements before implementation
• Reviewing vague language and acceptance criteria
• Assessing UX or API specification quality