swe-bench-lite

swe-bench-lite is a skill for Claude Code from greynewell/mcpbr. It costs 20 tokens per session (771 once invoked), scanned A, original, MIT.

A quick-start command for running SWE-bench Lite, a benchmark made of real software-engineering tasks used to evaluate coding agents. It runs sample tasks with mcpbr and saves JSON results plus a Markdown report.

In plain words
What is it for?
Run a small or full SWE-bench Lite evaluation, change the number of tasks, adjust verbosity, and choose where the results and report are saved.
Why use it?
It provides sensible defaults and checks the required Docker service, API key, and configuration file. This avoids assembling the benchmark command and output options by hand.

Skill for Claude Code

Written for Claude Code: a Claude Code plugin manifest. Also seen: positional $N argument.

Part of the mcpbr plugin — 3 skills shipped together

Good fit Run a small or full SWE-bench Lite evaluation, change the number of tasks, adjust verbosity, and choose where the results and report are saved.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/greynewell/mcpbr/benchmark-swe-lite
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add greynewell/mcpbr --skill benchmark-swe-lite
Clone the repo
git clone --depth 1 https://github.com/greynewell/mcpbr

Made for: Claude Code.

Or install mcpbr, the plugin that ships this one along with the rest of its 3 skills.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for swe-bench-lite

README.md
[![agentmods](https://agentmods.dev/badge/skills/greynewell/mcpbr/benchmark-swe-lite/github.svg)](https://agentmods.dev/skills/greynewell/mcpbr/benchmark-swe-lite)
Your own site
<a href="https://agentmods.dev/skills/greynewell/mcpbr/benchmark-swe-lite"><img src="https://agentmods.dev/badge/skills/greynewell/mcpbr/benchmark-swe-lite/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for swe-bench-lite

Your own site · 80×15
<a href="https://agentmods.dev/skills/greynewell/mcpbr/benchmark-swe-lite"><img src="https://agentmods.dev/badge/skills/greynewell/mcpbr/benchmark-swe-lite.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 20 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 771 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00020 $0.00771
Opus 5 $0.00010 $0.00385
Sonnet 5 $0.00004 $0.00154
Haiku 4.5 $0.00002 $0.00077

Measured 9d ago against content hash 7045de1ff87c, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade A, and why

swe-bench-lite scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude-plugin/skills/benchmark-swe-lite/SKILL.md · 103 lines

How it starts

The opening of the file, as written. The whole thing — 103 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Instructions

This skill provides a streamlined way to run the SWE-bench Lite benchmark with pre-configured defaults.

What This Skill Does

This skill runs a quick SWE-bench Lite evaluation with:

  • 5 sample tasks (configurable)
  • Verbose output for visibility
  • Results saved to results.json
  • Report saved to report.md

Prerequisites Check

Before running, verify:

  1. Docker is running:

    docker ps
    
  2. API key is set:

    echo $ANTHROPIC_API_KEY
    
  3. Config file exists:

    • Check for mcpbr.yaml in the current directory
    • If missing, run mcpbr init to generate it

Default Command

The default command for SWE-bench Lite:

mcpbr run -c mcpbr.yaml --dataset SWE-bench/SWE-bench_Lite -n 5 -v -o results.json -r report.md

Customization Options

Users can customize the run by modifying:

  • Sample size: Change -n 5 to any number (or remove for full dataset)
  • Config file: Change -c mcpbr.yaml to point to a different config
  • Verbosity: Use -vv for very verbose output
  • Output files: Change results.json and report.md to different paths

Example Variations

Minimal quick test (1 task)

mcpbr run -c mcpbr.yaml -n 1 -v

Full evaluation (all ~300 tasks)

mcpbr run -c mcpbr.yaml --dataset SWE-bench/SWE-bench_Lite -v -o results.json

MCP-only (skip baseline)

mcpbr run -c mcpbr.yaml -n 5 -M -v -o results.json

Specific tasks

mcpbr run -c mcpbr.yaml -t astropy__astropy-12907 -t django__django-11099 -v

Expected Runtime & Cost

For 5 tasks with default settings:

  • Runtime: 15-30 minutes (depends on task complexity)
  • Cost: $2-5 (depends on task complexity and model used)

What to Do If It Fails

  1. Docker not running: Start Docker Desktop
  2. API key missing: Set with export ANTHROPIC_API_KEY="sk-ant-..."
  3. Config missing: Run mcpbr init to generate default config
  4. Config invalid: Check that {workdir} placeholder is in the args array
  5. MCP server fails: Test the server command independently

Read the full file on GitHub · 103 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 9d ago First seen · 103 lines · 20 tokens per session scan A 7045de1ff87c

Subscribe to this mod's changes

swe-bench-lite is a skill published in the GitHub repository greynewell/mcpbr (10 stars, last pushed 4mo ago), licensed MIT. It adds 20 tokens to every session and 771 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.