ahmtsahin

Agent Context Bench — Security skill for Claude Code

Security community

Audit AGENTS.md, CLAUDE.md, Cursor and Copilot instructions; measure token cost and benchmark AI coding-agent context.

How to install Agent Context Bench

This entry records only its repository, not the path inside it, so there is no exact command to give. Open ahmtsahin/agent-context-bench and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agent Context Bench does

Audit AGENTS.md, CLAUDE.md, Cursor and Copilot instructions; measure token cost and benchmark AI coding-agent context.

Alternatives in Security

Anthropic Cybersecurity Skills — 734+ structured cybersecurity skills for AI agents · MITRE ATT&CK mapped · agentskills.io open standard · Work 3.8k ★
Pareto Mac — by ParetoSecurity - Serves as development guide for Mac security audit tool with build instructions, contribut 431 ★
Loongsuite Pilot — Local-first telemetry collector for AI coding agents — unified OpenTelemetry events for Claude Code, Codex, Cu 150 ★

README

agent-context-bench

[![npm version](https://img.shields.io/npm/v/agent-context-bench.svg)](https://www.npmjs.com/package/agent-context-bench) [![npm downloads](https://img.shields.io/npm/dm/agent-context-bench.svg)](https://www.npmjs.com/package/agent-context-bench) [![CI](https://github.com/ahmtsahin/agent-context-bench/actions/workflows/ci.yml/badge.svg)](https://github.com/ahmtsahin/agent-context-bench/actions/workflows/ci.yml) [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)

**Does your agent context help—or just add work?**

Audit `AGENTS.md`, `CLAUDE.md`, coding-agent `SKILL.md` files, Cursor rules, and Copilot instructions locally without LLM or API calls. Then A/B-test `none` vs `current` vs `optimized` context with your own agent, tasks, and verifier.

Agent Context Bench: a local static audit without LLM or API calls, followed by a benchmark that invokes your agent CLI

Static audit · no LLM/API calls | Benchmark · invokes your agent CLI | GitHub Action

Try it in 30 seconds

Requires Node.js 20 or newer.

npx agent-context-bench@latest .

Example audit output from a configured repository:

Status: Configured
Operational Score: 77/100
Inventory Score: 42/100
Context: 9 files, 924 lines, ~28,000 tokens

Static heuristic signals
- Outcome risk: moderate
- Token overhead risk: high
- Exploration overhead risk: high

Always loaded: 83/100 · 2 files · ~7,000 tokens
Deferred skills: 42/100 · 7 files · ~21,000 tokens

No usable context files (missing or empty)? The result is **Not configured**, not a perfect score. Use `-