What is AgentBench-Live CLAUDE.md?

Project instructions for AgentBench-Live, a tool that compares coding agents on tasks. They describe the project's current goals, planned work, testing setup, and deferred features.

Use them when maintaining AgentBench-Live, adding or updating agent adapters, running benchmark tasks in Docker, improving its comparison interface, or updating project documentation.

They give an agent the repository's priorities and working context before it changes code. This helps keep benchmark results, sandboxing, adapters, documentation, and the website aligned.

How jackjin1997/AgentBench-Live configures Claude Code

Reuse

Borrowing it

Nothing to install: this file belongs to jackjin1997/AgentBench-Live. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file

curl -O https://raw.githubusercontent.com/jackjin1997/AgentBench-Live/main/CLAUDE.md

Clone the repo

git clone --depth 1 https://github.com/jackjin1997/AgentBench-Live

Made for: Claude Code.

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.

README.md

[![agentmods](https://agentmods.dev/badge/instructions/jackjin1997/agentbench-live/claude-md/github.svg)](https://agentmods.dev/instructions/jackjin1997/agentbench-live/claude-md)

Your own site

<a href="https://agentmods.dev/instructions/jackjin1997/agentbench-live/claude-md"><img src="https://agentmods.dev/badge/instructions/jackjin1997/agentbench-live/claude-md/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.

Your own site · 80×15

<a href="https://agentmods.dev/instructions/jackjin1997/agentbench-live/claude-md"><img src="https://agentmods.dev/badge/instructions/jackjin1997/agentbench-live/claude-md.svg" alt="Reviewed on agentmods" width="80" height="20"></a>

Per session 536 This file is loaded in full into every session.

When invoked 536 The same file — it is already loaded in full.

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.

Model	Per session	Once invoked
Fable 5.1	$0.00536	$0.00536
Opus 5.5	$0.00214	$0.00214
Sonnet 5.5	$0.00107	$0.00107
Haiku 4.5	$0.00054	$0.00054

Measured 2mo ago against content hash 5cb1cc85dfe9, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

AgentBench-Live CLAUDE.md 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 2mo 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.md · 56 lines

How it starts

The opening of the file, as written. The whole thing — 56 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AgentBench-Live

Project Goal

Get on GitHub Trending. Current: 2 stars. Need: 50-150+ stars/day spike.

Strategy: Credibility First, Then Ride the Wave

Fix trust gaps (Docker sandbox, more agents, real scores)
Wait for a major agent launch, benchmark it day-1
Coordinated multi-channel social push (Twitter, Reddit, HN, 小红书, V2EX)

Current Sprint Scope (accepted from CEO review 2026-03-19)

Docker sandbox (replace tempdir with real isolation)
Add Codex CLI adapter (already existed)
Add Aider adapter
Re-run ALL agents × ALL tasks with Docker (get real scores) ← USER ACTION NEEDED
Methodology doc (docs/methodology.md)
Frontend redesign (radar charts, domain filtering, hero section)
Social card generator (shareable comparison images)
Update README for 4 agents

Deferred (TODOS)

Agent Output Viewer (show actual agent work per task) — P2
GitHub Actions automated benchmark CI — P2
Automated nightly benchmark runs — Phase 2

Architecture

CLI (click) → Runner → Adapter Registry → SandboxFactory
                                            ├── DockerSandbox (primary)
                                            └── LocalSandbox (fallback)
                                         → Evaluator (auto + LLM judge)
                                         → Ranking → Leaderboard (GitHub Pages)

Key Technical Decisions

4 agents: Claude Code, Gemini CLI, Codex CLI, Aider
Docker sandbox with fallback to local tempdir
SandboxFactory pattern (not inheritance replacement)
1 trial for initial run, pass@k requires multiple trials
Sanitize API keys from logs/output

Critical Gaps to Fix

Docker not available → fall back to LocalSandbox gracefully
Agent timeout → record score=0 and continue, don't crash
LLM judge failures → fall back to auto-eval, don't crash

Competitors

SWE-bench: ~4k stars (academic, GitHub issues)
PinchBench: ~640 stars (company-backed)
ai-coding-lang-bench: ~98 stars (language comparison)
TheAgentCompany: ~659 stars (simulated company)

Read the full file on GitHub · 56 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.

2mo ago First seen · 56 lines · 536 tokens per session scan A 5cb1cc85dfe9

Subscribe to this mod's changes

AgentBench-Live CLAUDE.md is an instructions file published in the GitHub repository jackjin1997/AgentBench-Live (4 stars, last pushed 5mo ago), licensed MIT. It adds 536 tokens to every session, about $0.0021 per session on Opus 5.5. 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.

Claude Code Evaluation agent-benchmark ai-agents benchmark cli-agents

vscode buildNext.instructions.md

Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).

microsoft/vscode · 9,715 tokens

AgentBench-Live: Instructions file for Claude Code