Claude Skills by yulonglin

Process bulk edits from a comments/feedback file safely using sequential agents.

Delegate tasks to Claude Code CLI for a Claude-powered second opinion, parallel implementation, or plan review. Use when you want Claude (not Codex/Gemini) as a delegate — e.g., tasks needing Claude's judgment, tool use, or MCP access. Do NOT use for: - Pure implementation with clear specs (use codex-cli) - Large context analysis (use gemini-cli) - Quick edits under ~10 lines (faster to do directly)

Delegate well-scoped implementation tasks to Codex CLI. Use when: - Implementing a well-defined function, module, or feature - Bug fixes with known root cause and clear fix - Scoped refactoring (rename, extract, restructure within a module) - File generation from a template or spec - Getting a second opinion on an implementation plan or approach Do NOT use for: - Exploration or codebase understanding (use Explore agent) - Multi-turn refinement or ambiguous tasks (do it yourself) - Quick edit...

Remove AI generated code slop from the current working directory.

Fix all merge conflicts on the current Git branch non-interactively and make the repo buildable and tested.

Use when running API-heavy experiments (LLM evals, batch processing). Covers memory-efficient patterns, batch APIs vs async, and avoiding the parallel process anti-pattern.

Set up automated experiment logging using Hydra for configuration management and Inspect AI for evaluations.

Generate a research spec in the repo from a template if it doesn't exist.

Set up and use SLURM helpers for the MATS compute cluster with A100 and L40 GPUs.

Analyze research papers and technical articles with critical rigor. Use when (1) reading PDFs/arxiv papers, (2) doing literature reviews, (3) evaluating methodology/claims, (4) comparing approaches across papers, (5) user shares a paper URL/file, or (6) learning concepts that require reading papers (e.g. "explain GRPO", "what is constitutional AI", "how does deliberative alignment work"). Provides grounded analysis with explicit citations (never fabricates), assesses evidence strength, and id...

Generate reproducibility reports for LLM/ML experiments. Use when (1) completing an experiment, (2) sharing results with collaborators, (3) preparing for publication, (4) archiving experiment runs, or (5) user asks for reproducibility documentation. Auto-extracts from code/configs/Hydra, asks for gaps.

Run experiments with output isolation to prevent context pollution.

Interview-based spec development for research experiments. Iteratively questions user about hypotheses, variables, baselines, and resources before writing a complete research specification.

Create Anthropic-style TikZ diagrams for ML/AI safety papers. 17 patterns, semantic colors, NeurIPS-ready.

Orchestrate agent teams for parallelizable multi-faceted tasks. Use when work requires inter-agent communication, competing hypotheses, or multi-file parallel implementation.

Compact current work with optional focus using agent-context-summariser

Externalise important parts of the current conversation into an .md file for handover.

Generate Claude Code usage insights reports. Analyzes session transcripts, extracts structured facets via Gemini CLI, and produces coaching-style HTML reports with temporal trends and actionable recommendations. Use when: - You want to understand your Claude Code usage patterns - You want a report filtered to a specific project - You want to see how your usage has changed over time - You want to list all projects with session counts Requires: Python 3.9+, Gemini CLI (`brew install gemini-cli`)

Write clearly without LLM-isms. Reference when drafting papers, blogs, or professional communications.

Identify and fix slides with content overflow or blank pages in Slidev presentations.

DEPRECATED: Use /review-draft --critics=humanizer instead. LLM cliche detection is now integrated into review-draft.

Guide for structuring research presentations, slides, and updates.

Comprehensive writing review using specialized critics (clarity, narrative, facts, red-team).

This skill should be used when the user asks to "review a paper", "critique a manuscript", "give feedback on a draft", "review this write-up", "analyze paper quality", or wants constructive feedback on ML/AI research writing based on Neel Nanda's paper-writing criteria.

Slidev presentation framework helper. Use when creating slides, fixing Slidev issues, exporting to PDF, or when user mentions Slidev, slides.md, presentation export, or slide overflow.

Guide for effective communication in situations requiring negotiation, persuasion, or strategic thinking. Built on Chris Voss's FBI negotiation approach - tactical empathy and emotion labeling as primary tools. Use when drafting messages for rentals, salary discussions, work requests, declining offers, or any situation where you need to be both warm and strategically clear.

Run custom usage analytics on all Claude Code sessions

Applies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. Taken from https://github.com/anthropics/skills/blob/main/skills/brand-guidelines/SKILL.md

This skill should be used when the user asks to "commit and push", "commit push", "sync changes", "push changes", "commit and sync", or "update remote". Handles the full workflow of committing changes, pulling with rebase, and pushing to remote.

Commit current work with optional message. Handles git status/diff checking and message formatting.

Modern CLI tool mappings (eza, fd, rg, bat, dust, duf, fzf, zoxide, delta, jq). Use for file listing, searching, and disk usage.

Interview-based spec development for large features. Iteratively questions user about implementation, UX, concerns, and tradeoffs before writing a complete specification.