Claude Skills
The open format is called Agent Skills and works in Claude Code, Codex, Cursor and other agents — most people know it as Claude Skills.
Every Agent Skill we could find on GitHub, deduplicated by content. 79 566 files from 1 758 authors, of which 61 913 are unique — the rest is the same skill repackaged into someone else's repository. For each one: what it weighs in tokens, whether it ships runnable scripts, and which MCP servers it needs.
How to write lookup formulas (INDEX/MATCH, VLOOKUP, SUMPRODUCT-based lookups) that work in both Excel and Gnumeric/LibreOffice for multi-condition lookups. Use this when filling cells with formulas that reference data based on multiple criteria.
How to calculate net exports as % of GDP, descriptive statistics, and weighted mean for GCC countries in the gdp.xlsx workbook. Use this for the specific task of filling in the Task sheet with lookup formulas, percentage calculations, and SUMPRODUCT weighted mean.
How to inspect the structure of an Excel workbook before writing formulas, including checking sheet names, cell values, data types, and layout. Use this whenever you need to understand the actual content of an Excel file before modifying it.
Provides the official Anthropic brand color palette and typography tokens. Use this skill whenever you need canonical HEX values for Anthropic brand colors or the correct heading font name for any Anthropic design artifact.
Use this skill to look up the four canonical tonal patterns for seven-character regulated verse (七律 Qilv). Provides exact 平/仄 sequences for all 8 lines under each of the four standard forms, plus the rule for checking positions 2, 4, 6.
Use this skill to fill in the California SC-100 Small Claims Court PDF form with case data and save the filled PDF. Handles text fields, checkboxes, and radio buttons using pypdf. Run the inspection skill first to confirm field names, then use this skill to write the filled PDF to /root/sc100-filled.pdf.
Use this skill to inspect all form fields in a PDF file, printing their names, types, and current values/export values. Useful for understanding the structure of a PDF form before filling it.
How to run DBSCAN with a custom anisotropic distance metric using sklearn, including defining callable metrics and computing cluster centroids from labeled output.
How to perform greedy nearest-neighbor matching between two sets of 2D points (cluster centroids vs expert annotations) with a maximum distance threshold, returning matched pairs for F1/delta computation.
How to compute the Pareto frontier from a set of (F1, delta) optimization results where F1 is maximized and delta is minimized. Includes filtering, dominance checking, and CSV output formatting.
Use this skill first to locate the trivy binary and its pre-downloaded vulnerability database cache on the system before running any security audit.
Use this skill to run Trivy in offline mode against /root/package-lock.json and produce the security audit CSV at /root/security_audit.csv. Use only after locating the trivy cache directory.
Find the earthquake within the Pacific plate (Code == "PA") that is farthest from the Pacific plate boundary lines in PB2002_boundaries.json. Uses an equal-area projection for accurate distance measurement. Filters boundaries where PlateA or PlateB equals "PA".
Load GeoJSON files with GeoPandas and inspect their structure — columns, CRS, geometry types, and sample rows. Use this before any spatial analysis to understand field names and data layout.
Count the number of distinct stocks (equity positions only) held by Renaissance Technologies. Filters to SH (share) type only and excludes options/derivatives by checking PUTCALL column. Use this specifically for Q2 answer about Renaissance stock count.
Load the INFOTABLE (holdings) parquet for a given quarter and filter by accession_number to get a specific fund's holdings. Handles column name variations. Use this after obtaining an accession_number from the coverpage.
Load and search the COVERPAGE parquet file for a specific fund manager using fuzzy matching on the FILINGMANAGER_NAME field. Returns the best matching row including accession_number, AUM, and other fund details. Use this when you need to find a fund's accession number by name.
Use `gh search issues` to fetch issues from a specific repo within a date range. Returns structured JSON. More reliable than `gh issue list --search` for date-range filtering.
Use `gh search prs` to fetch pull requests from a specific repo within a date range. Returns structured JSON for downstream processing. More reliable than `gh pr list --search` for date-range filtering.
Compile PR and issue metrics into a report.json file at /app/report.json using the required schema for the December community pulse write-up.
Use this skill when you need to find, install, or configure a Python 3.10 environment on the system. Checks for existing python3.10 binary, installs if missing, and sets up pip for that version.
Use this skill to implement the simpo_loss function in SimPOTrainer based on the SimPO paper. The loss combines a length-normalized reward with a margin gamma and uses BCE loss without a reference model.
Use this skill to install all required packages for the SimPO project using Python 3.10 specifically. Installs torch, transformers, trl, and other dependencies into the python3.10 environment.
Use this skill to execute the SimPO unit test with Python 3.10, save the loss results to /root/loss.npz, and log Python version and package info to /root/python_info.txt using the correct Python 3.10 executable.
Use this skill to classify and move 100+ PDF/PPTX/DOCX files into 5 subject folders (LLM, trapped_ion_and_qc, black_hole, DNA, music_history) based on content analysis. Handles keyword scoring with filename tiebreaking and music_history as catch-all default.
Use when translating Python OOP/functional code to idiomatic Scala 2.13, covering class hierarchies, enumerations, error handling, and naming conventions
Use when implementing tokenizer abstractions in Scala 2.13 — covers Token/TokenType ADTs, tokenizer traits, builder pattern, and batch processing conventions
Use this skill to read and parse structured email data from a JSON file, extracting fields like message ID, sender email, subject, body, and meeting constraints (duration, preferred times/dates).
Use when building a D3.js v6 bubble chart with force simulation, sector clustering, collision detection, tooltips, and interactive highlighting. Handles ETFs (no marketCap/sector) as a separate cluster.
Use when building an HTML data table that syncs highlighting with a D3 bubble chart on click interactions in both directions.
Use when you need to create output directory structures and copy files into them for web app deployment.
Use when you need to obtain the D3.js v6 minified library file and save it locally for a web app.
Use when you need to enumerate files in a directory, optionally filtering by extension. Returns sorted list of file paths.
Use when you need to read and parse CSV files from the filesystem in Python. Handles encoding, missing values, and returns structured data.
Use when generating all files for a single-page web app: index.html, CSS, and JS. Ensures correct relative paths, proper script loading order, and valid HTML structure.
Use this skill to understand the GLM (General Lake Model) configuration for Lake Mendota, including how to read and modify the glm3.nml file, what parameters are allowed to change, and how the model is structured. Apply this before running or calibrating GLM.
Use this skill to correctly parse GLM NetCDF output (output.nc) into a DataFrame of (datetime, depth, temperature) triples. Handles the 4D variable shape (ntime, nlayer, 1, 1), correct depth conversion using fixed lake_depth from glm3.nml, and datetime reconstruction using manual timedelta from a fixed reference date.
Use this skill to convert a completed day-by-day travel plan into the required JSON output format and write it to /app/output/itinerary.json. Enforces field rules, naming conventions, and completeness checks.
Use this skill to build a multi-day travel itinerary from structured requirements (origin, cities, dates, budget, constraints). It queries available database files and produces a day-by-day plan with transportation, meals, attractions, and accommodations.
Use this skill to validate a completed itinerary JSON against all task requirements before finalizing output. Run this checklist after formatting and before writing the file.
Convert all extracted keyframe PNG images from RGB to grayscale inplace, overwriting the original files. Also convert template images (coin.png, enemy.png, turtle.png) to grayscale. Use this before running object counting.
Count coins, enemies, and turtles in each keyframe using the provided count_objects.py script with threshold=0.9 and dedup_min_dist=3. Parse stdout for integer counts. Write results to /root/counting_results.csv.
Extract key frames from an MP4 video file and save them as PNG images in a target directory. Use this skill when you need to convert a video into individual frame images for further analysis.
Use when you need to retrieve data from a table using two simultaneous conditions (e.g., row label + column header). Covers INDEX&MATCH, XLOOKUP&MATCH, VLOOKUP&MATCH, and HLOOKUP&MATCH patterns in Excel.
Use when calculating net exports as a percentage of GDP in Excel, including computing summary statistics (min, max, median, mean, percentiles) and a GDP-weighted mean using SUMPRODUCT. Covers formula structure, rounding conventions, and display formatting.
Use when writing formulas in one Excel sheet that reference data in another sheet. Covers syntax, absolute vs relative references, and best practices for multi-sheet lookup formulas.
Create technical diagrams and posters adhering to Anthropic's official brand guidelines. Use this skill to produce exploded-view hardware visualizations with specific brand color tokens, typography, and minimalist styling.
Rules for determining the Ping (平) or Ze (仄) status of a character based on modern Mandarin Pinyin tones.
Composing a Chinese Seven-Character Regulated Verse (七言律诗) following strict traditional structural, tonal, and rhyming rules using modern Mandarin.
Fills a PDF form with provided data. It maps field keys to their respective pages and applies values. It handles multi-line text areas by assigning specific strings to the identified sequential keys. For checkboxes, it uses the precise export values found during inspection. All dates must be formatted as 'xxxx-xx-xx'.
Splits a long string of text (like a claim description or reason) into multiple parts that fit into sequential PDF field keys. Use this when the SC-100 form provides multiple lines (e.g., 'Reason_Line1', 'Reason_Line2') for a single explanation.
Extracts detailed information about form fields in a PDF, including field keys, types, page indices, current values, and allowed export values for checkboxes and radio buttons. Use this to identify how multi-line text is split across different keys (e.g., 'Reason_1', 'Reason_2') and to find the exact string required to check a box (e.g., 'Yes', '1', or 'On').
Implementing a weighted Euclidean distance metric for DBSCAN to account for anisotropic spatial features.
Matching predicted cluster centroids to ground truth points using a greedy distance-based approach to calculate F1 score and precision/recall.
Running a multi-parameter grid search efficiently using parallel processing to find optimal hyperparameters.
Organizing annotation data by image identifiers to prepare for image-by-image clustering analysis.
Identifying the set of non-dominated solutions in a multi-objective optimization problem.
Formatting security audit findings into a standardized CSV format for reporting and compliance.
Claude Skills — questions
Answers built from the skills we actually parsed.