Research

Use when generating multiple curriculum-aligned concept files in parallel (5+ concepts) - researches curriculum, creates concept list, spawns parallel generation agents, orchestrates review loop until all pass

Synthesizes research findings into coherent narratives with uncertainty quantification. Use when integrating findings from multiple sources, creating research summaries, drawing conclusions from evidence, or communicating research results. Triggers on phrases like "synthesize", "integrate findings", "what's the conclusion", "summarize research", "overall picture", "bring together".

Document complex domain knowledge as self-contained packages with multiple reading paths

Invoke deep sequential thinking for complex problem-solving. Use when the user says 'use ultrathink', 'ultrathink', or when tackling problems that require careful step-by-step reasoning, planning, hypothesis generation, or multi-step analysis.

Add project repository links to PROJECTS.md when user mentions completing, starting, or sharing a hands-on project. Use when user references external repos or implementation work.

Use after feature-implement completes - performs final quality check with feature-research from fresh context, identifies issues, offers fix/loop/document choices, updates Z01/Z02 if implementation deviated from plan

Use when implementing features from Jira tickets, PRDs, or user requirements. Orchestrates Research-Plan-Implement workflow with quality gates for hallucination, overengineering, and underengineering detection.

Verify citations and factual claims in long-form content. Checks URLs for accessibility and accuracy, validates quotes, analyzes source credibility, and distinguishes facts from opinions. Use when fact-checking article drafts, validating citations, or reviewing claims before publication.

This skill manages experimental reproducibility, pooling, and consensus strategies. This skill operates in two distinct modes based on the input state. (1) Pre-Peak Calling (BAM Mode): It merges all BAMs, generate the merge BAM file to prepare for track generation and (if provided with >3 biological replicates) splits them into 2 balanced "pseudo-replicates" to prepare for peak calling. (2) Post-Peak Calling (Peak Mode): If provided with peak files (only support two replicates, derived from either 2 true replicates or 2 pseudo-replicates), it performs IDR (Irreproducible Discovery Rate) analysis, filters non-reproducible peaks, and generates a final "conservative" or "optimal" consensus peak set. Trigger this skill when you need to handle more than two replicates (creating pseudo-reps) OR when you need to merge peak lists.

When asked implementation questions or tool selection questions like "how do I implement", "what's the best way to", "how should I", "which library", "what tool should I use", or "how might we", search GitHub for prior art, code examples, and proven approaches before proposing solutions

Supports Unity C# script editing, searching, and refactoring. Enables TDD cycle code editing, symbol navigation, reference searching, and structured editing. Use when: C# editing, script search, symbol search, refactoring, code indexing, class creation, method addition

Run Stanford STORM (knowledge-storm) to generate comprehensive, Wikipedia-style articles with citations. Requires LLM and Search API keys (Bing or You.com).

Use when encountering any bug, test failure, or unexpected behavior (including race conditions, deadlocks, concurrency issues) - four-phase framework (root cause investigation, pattern analysis, hypothesis testing, implementation) with specialized techniques for deep call stack tracing and concurrency debugging

This skill should be used when the user asks to "分析PDF", "解读文档", "分析PDF文件", "PDF解读", "extract content from PDF", "analyze PDF document", "parse academic paper", or provides a PDF file path for content extraction and analysis. Provides objective PDF document analysis and content extraction for professional WeChat content creation.

Conducts deep company research and creates a dossier to support applications. Use before applying, interviewing, or for cultural fit assessment.

Use when the user asks to run Codex CLI (codex exec, codex resume) or references OpenAI Codex for code analysis, refactoring, or automated editing

Use when user provides YouTube URL and says "document this", "create notes", or "save this video". Automatically extracts transcript, determines folder placement, and generates Knowledge Framework documentation with MECE structure, Mermaid diagrams, and clickable timestamp citations.

Hypothesis-driven hunt plan for suspicious PowerShell, plus query snippets for common telemetry.

Scaffold complete forecasting experiments with Nixtla libraries - creates config files, experiment harness, and evaluation workflows

Use when the user asks to run Codex CLI (codex exec, codex resume) or references OpenAI Codex for code analysis, refactoring, or automated editing