# benchflow-ai/skillsbench

SkillsBench evaluates how well skills work and how effective agents are at using them.

Repository: https://github.com/benchflow-ai/skillsbench
Canonical: https://ross.abutalabs.com/products/skillsbench
Homepage: https://www.skillsbench.ai
Language: PDDL
License: Apache-2.0
License Family: permissive
Topics: benchmark, skillsbench
Last push: 2026-07-23T19:09:35+00:00

## Health v2 (maintenance only)
Score: 76/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 94, release rhythm 88, longevity 17
- inputs: {"age_days": 247, "days_push": 41, "days_rel": 80, "gap_med": 2, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1725, forks 364 (observed 2026-08-28T04:05:28.095091+00:00)

## What it is
SkillsBench is the first benchmark for evaluating how effectively AI agents use skills—modular folders of instructions, scripts, and resources—across expert-curated professional tasks. It provides sandboxed task packages with oracle solutions and outcome-based verifiers, runnable via the BenchFlow CLI on Modal or Docker.

## Use cases
- benchmark how well AI agents use skills
- compare LLM performance with and without skills
- evaluate agent skill composition across tasks
- measure agent resolution rate and wall-clock efficiency
- create and contribute agent evaluation tasks
- run reproducible agent evaluations in cloud sandboxes

## When to choose
- you need to measure how effectively agents leverage skills on specialized workflows
- you want to compare frontier models on skill-using agent tasks
- you need a gym-style evaluation harness with oracle sanity checks and verifiers
- you want to contribute or curate benchmark tasks for agent skills

## When to avoid
- you need a general-purpose agent framework for building products rather than evaluation
- you want a simple unit-testing tool for code
- you cannot use Modal or Docker sandbox execution
- you need benchmarking of non-agent software performance

## Facets
- artifact type: dataset
- maturity: active
- function: benchmarking, agent-framework, testing
- domain: artificial-intelligence, large-language-models, developer-tools, tutorials
- platform: python, cli, cloud
- tags: agent-skills, benchmark, llm-evaluation, skill-composition, sandboxed-tasks, leaderboard, benchflow, evaluation, ai-agents, docker

## Member repositories
- benchflow-ai/skillsbench (main) score 76

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:28.095091+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:32:34.895333+00:00, confidence not recorded.
  - readme: https://github.com/benchflow-ai/skillsbench (fetched 2026-08-28T04:05:28.095091+00:00, sha 3e4444821f0a)
  - homepage: https://www.skillsbench.ai (fetched 2026-08-29T11:09:09.799238+00:00, sha efc21c281b6b)
  - site_page: https://www.skillsbench.ai/docs/getting-started (fetched 2026-08-29T11:09:09.808806+00:00, sha e477c5e677d9)
  - site_page: https://www.skillsbench.ai/docs (fetched 2026-08-29T11:09:09.810860+00:00, sha 2ac6d9bc8be6)
- Data as of 2026-08-30T08:39:29.467469+00:00.