Datasets:

GeniusHTX
/

SWE-Skills-Bench

skill_id stringlengths 3 31	name stringlengths 12 35	description stringclasses 2 values	type stringclasses 5 values	task_prompt stringlengths 1.01k 2.23k	skill_document stringlengths 495 26.8k	test_code stringlengths 3.04k 18.1k	repo_url stringlengths 0 72	repo_commit stringclasses 2 values	docker_image stringclasses 8 values
add-uint-support	Add UInt Support	Restore uint32/uint64 operator support in PyTorch	repair	# Task: Enable Unsigned Integer Support for Target Operators ## Background Several operators in PyTorch do not currently support unsigned integer types (uint16, uint32, uint64). When users attempt to perform calculations with these tensor types, the system returns an error stating that the type is not implemented. M...	--- name: add-uint-support description: Add unsigned integer (uint) type support to PyTorch operators by updating AT_DISPATCH macros. Use when adding support for uint16, uint32, uint64 types to operators, kernels, or when user mentions enabling unsigned types, barebones unsigned types, or uint support. --- # Add Unsig...	""" Unit Test for UInt32/64 Operator Support in PyTorch """ import torch import pytest class TestUIntOperators: """Tests for uint32 and uint64 operator support.""" @pytest.fixture(params=["uint32", "uint64"]) def dtype(self, request): """Parametrized fixture: uint32 and uint64.""" dtype_...			zhangyiiiiii/swe-skills-bench-pytorch:latest
fix	React Code Fix & Linter	See task file for detailed mission requirements.	fix	# Task: Fix ESLint Violations in TypeScript Codebase ## Background The upgradle project uses TypeScript + ESLint for code quality enforcement. Currently, the `src/` directory contains multiple ESLint rule violations that need to be addressed: - `no-unused-vars` - `@typescript-eslint/no-explicit-any` - `eqeqeq` (stri...	--- name: fix description: Use when you have lint errors, formatting issues, or before committing code to ensure it passes CI. --- # Fix Lint and Formatting ## Instructions 1. Run `yarn prettier` to fix formatting 2. Run `yarn linc` to check for remaining lint issues 3. Report any remaining manual fixes needed ## C...	""" Test for 'fix' skill — React Code Fix & Linter Validates that the Agent scanned and fixed all ESLint violations in the upgradle TypeScript codebase so that `npm run lint` passes cleanly. """ import os import subprocess import glob import re import pytest from _dependency_utils import ensure_npm_dependencies @py...	https://github.com/michaelasper/upgradle	5f292188e9b427a96d3573b29e5677e4cdce58ea	zhangyiiiiii/swe-skills-bench-python
tdd-workflow	TDD Workflow	See task file for detailed mission requirements.	feature	# Task: Implement Smart Coupon Calculator ## Required File Paths (Agent must only modify/create under these) - MUST modify: `src/calculator.py` — implement or update `SmartCouponCalculator` here. ## Background We need a flexible discount calculation system for our e-commerce platform that can handle multiple promo...	--- name: tdd-workflow description: Use this skill when writing new features, fixing bugs, or refactoring code. Enforces test-driven development with 80%+ coverage including unit, integration, and E2E tests. --- # 測試驅動開發工作流程此技能確保所有程式碼開發遵循 TDD 原則，並具有完整的測試覆蓋率。 ## 何時啟用 - 撰寫新功能或功能性程式碼 - 修復 Bug 或問題 - 重構現有程式碼 - 新增 API 端...	""" Test for 'tdd-workflow' skill — Smart Coupon Calculator Validates that the Agent implemented SmartCouponCalculator with progressive, category, and user-tier discounts in src/calculator.py. """ import os import sys import importlib import subprocess import pytest class TestTddWorkflow: """Verify SmartCouponCa...	https://github.com/tdd-starters/python		zhangyiiiiii/swe-skills-bench-python
security-review	Security Review (zh-TW)	See task file for detailed mission requirements.	feature	# Task: Implement Secure Export API Endpoints for BabyBuddy ## Background We need to add export endpoints to BabyBuddy's REST API that allow users to export feeding and sleep records. The implementation must enforce proper authentication and authorization to ensure users can only access their own children's data. ##...	--- name: security-review description: Use this skill when adding authentication, handling user input, working with secrets, creating API endpoints, or implementing payment/sensitive features. Provides comprehensive security checklist and patterns. --- # 安全性審查技能此技能確保所有程式碼遵循安全性最佳實務並識別潛在漏洞。 ## 何時啟用 - 實作認證或授權 - 處理使用者...	""" Test for 'security-review' skill — Secure Export API for BabyBuddy Validates that the Agent implemented authenticated, authorized export endpoints with proper serializers, views, URLs, and security checks. """ import os import ast import subprocess import pytest from _dependency_utils import ensure_python_depende...	https://github.com/babybuddy/babybuddy		zhangyiiiiii/swe-skills-bench-python
springboot-tdd	Spring Boot TDD	See task file for detailed mission requirements.	feature	# Task: Add Pet Weight Tracking Feature to PetClinic ## Background We need to add a weight tracking feature to the Spring PetClinic application. Pet owners should be able to record and view their pets' weight history over time. ## Files to Create/Modify - `src/main/java/org/springframework/samples/petclinic/owner/W...	--- name: springboot-tdd description: Test-driven development for Spring Boot using JUnit 5, Mockito, MockMvc, Testcontainers, and JaCoCo. Use when adding features, fixing bugs, or refactoring. --- # Spring Boot TDD Workflow TDD guidance for Spring Boot services with 80%+ coverage (unit + integration). ## When to Us...	""" Test for 'springboot-tdd' skill — Spring Boot TDD Workflow Validates that the Agent added REST endpoints with TDD approach in the Spring PetClinic application: controller, service, model, and tests. """ import os import subprocess import pytest class TestSpringbootTdd: """Verify Spring Boot TDD implementatio...	https://github.com/spring-projects/spring-petclinic		zhangyiiiiii/swe-skills-bench-jvm
add-admin-api-endpoint	Ghost Admin API Endpoint Creator	See task file for detailed mission requirements.	feature	# Task: Create audit_logs Admin API Endpoint for Ghost CMS ## Background We need to add an `audit_logs` resource endpoint to the Ghost Admin API, allowing administrators to query recent user operation records for security and compliance purposes. ## Files to Create/Modify * `ghost/core/core/server/api/endpoints/aud...	--- name: Add Admin API Endpoint description: Add a new endpoint or endpoints to Ghost's Admin API at `ghost/api/admin/**`. --- # Create Admin API Endpoint ## Instructions 1. If creating an endpoint for an entirely new resource, create a new endpoint file in `ghost/core/core/server/api/endpoints/`. Otherwise, locate...	""" Test for 'add-admin-api-endpoint' skill — Ghost Admin API Endpoint Validates that the Agent added a new audit_logs Admin API endpoint in Ghost with model, endpoint handler, route registration, and tests. """ import os import re import subprocess import pytest from _dependency_utils import ensure_npm_dependencies ...	https://github.com/TryGhost/Ghost		zhangyiiiiii/swe-skills-bench-python
mcp-builder	MCP Server Builder	See task file for detailed mission requirements.	feature	# Task: Build MCP Server for Markdown Knowledge Base with SQLite ## Background We need to create a hybrid MCP (Model Context Protocol) server using TypeScript and `@modelcontextprotocol/sdk` that connects a local Markdown knowledge base with a SQLite metadata database. ## Files to Create/Modify - `src/markdown-sqli...	--- name: mcp-builder description: Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). license...	""" Test for 'mcp-builder' skill — MCP Server Builder Validates that the Agent created a new MCP (Model Context Protocol) server implementation with TypeScript source, build config, and tests. """ import os import subprocess import pytest from _dependency_utils import ensure_npm_dependencies @pytest.fixture(scope="...	https://github.com/modelcontextprotocol/servers		zhangyiiiiii/swe-skills-bench-python
python-resilience	Python Resilience Patterns	See task file for detailed mission requirements.	feature	# Task: Implement Resilient Transport Layer for httpx ## Background We need to add a resilient transport module to the httpx library that provides automatic retry and circuit breaker capabilities directly within the httpx transport layer. ## Files to Create/Modify - `httpx/_transports/resilient.py` - Resilient tran...	--- name: python-resilience description: Python resilience patterns including automatic retries, exponential backoff, timeouts, and fault-tolerant decorators. Use when adding retry logic, implementing timeouts, building fault-tolerant services, or handling transient failures. --- # Python Resilience Patterns Build fa...	""" Test for 'python-resilience' skill — Resilient Transport Layer for httpx Validates that the Agent implemented ResilientTransport with retry and circuit-breaker logic in httpx/_transports/resilient.py. """ import os import sys import ast import subprocess import importlib import pytest class TestPythonResilience:...	https://github.com/encode/httpx		zhangyiiiiii/swe-skills-bench-python
xlsx	Excel & Spreadsheet Automation	See task file for detailed mission requirements.	feature	# Task: Implement Sales Report Generation Engine for openpyxl ## Background We need to add a report generation engine to the openpyxl library that can produce Excel reports with automated summary formulas, conditional formatting, and trend charts. ## Files to Create/Modify - `openpyxl/utils/report_engine.py` - Repo...	--- name: xlsx description: "Use this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new s...	""" Test for 'xlsx' skill — Excel & Spreadsheet Automation Validates that the Agent implemented generate_sales_report() in openpyxl/utils/report_engine.py with summary formulas, conditional formatting, and trend charts. """ import os import sys import ast import subprocess import tempfile import pytest class TestXls...	https://github.com/ericgazoni/openpyxl		zhangyiiiiii/swe-skills-bench-python
turborepo	Turborepo Monorepo Build System	See task file for detailed mission requirements.	feature	"# Task: Create Turborepo Monorepo Example with Cache Demonstration\n\n## Background\n\nWe need a co(...TRUNCATED)	"---\nname: turborepo\ndescription: \|\n Turborepo monorepo build system guidance. Triggers on: turb(...TRUNCATED)	"\"\"\"\nTest for 'turborepo' skill — Turborepo Monorepo Configuration\nValidates that the Agent s(...TRUNCATED)	https://github.com/vercel/turbo		zhangyiiiiii/swe-skills-bench-python

End of preview. Expand in Data Studio

Dataset Summary

SWE-Skills-Bench is a benchmark dataset for evaluating whether injected skill documents — structured packages of procedural knowledge — measurably improve LLM agent performance on real-world software engineering tasks.

The dataset contains 49 skills spanning 565 task instances across six software engineering domains (Deployment & DevOps, Analytics & Monitoring, API Development, Data Science & ML, Security & Testing, and Developer Tools). Each skill is grounded in an authentic GitHub repository at a fixed commit, paired with a curated skill document and a deterministic pytest test suite that encodes the task's acceptance criteria.

The dataset is designed to answer: Does giving an agent a skill document actually help? The primary evaluation metric is pytest pass rate, measured under two conditions — with and without skill injection — to compute a pass-rate delta (ΔP) per skill.

The dataset was released as part of SWE-Skills-Bench: Do Agent Skills Actually Help in Real-World Software Engineering?

Dataset Structure

An example of a SWE-Skills-Bench datum is as follows:

skill_id:        (str) - Unique skill identifier, e.g. "fix", "tdd-workflow".
name:            (str) - Human-readable task name.
description:     (str) - One-line description of the task.
type:            (str) - Task category, e.g. "repair", "feature", "fix".
task_prompt:     (str) - Full task prompt passed to the agent (Markdown).
skill_document:  (str) - Curated skill document injected as agent context (Markdown).
test_code:       (str) - Pytest test suite used to evaluate the agent's output.
repo_url:        (str) - Target GitHub repository URL.
repo_commit:     (str) - Fixed commit hash for reproducibility.
docker_image:    (str) - Pre-configured Docker image for the evaluation environment.

Supported Tasks

SWE-Skills-Bench proposes a paired evaluation task: given a task prompt (with or without an injected skill document), an agent must complete a software engineering task on a real codebase. Correctness is verified by running the associated pytest test suite inside a Docker container.

Downloads last month: 31

Size of downloaded dataset files:

982 kB

Size of the auto-converted Parquet files:

380 kB

Number of rows:

Paper for GeniusHTX/SWE-Skills-Bench

SWE-Skills-Bench: Do Agent Skills Actually Help in Real-World Software Engineering?

Paper • 2603.15401 • Published 24 days ago • 18