Skill Security Audit Test Fixture
Provides a deliberately flawed mock skill fixture to test and validate Claude Code skill security auditing tools.
Designed specifically as a synthetic test fixture within the ai-dev-toolkit, this resource simulates dangerous behaviors and anti-patterns—such as credential exfiltration, prompt injection phrasing, and out-of-scope filesystem modifications—to evaluate and benchmark Claude Code security auditing capabilities like audit-skills. It serves as a regression and verification benchmark for AI skill linters to ensure unauthorized command execution and behavioral anomalies are accurately flagged.
主な機能
01Provides structured baseline findings to benchmark security audit rules
02Tests out-of-scope filesystem write detection across home directory boundaries
03Emulates command patterns that trigger unauthorized data exfiltration alerts
041 GitHub stars
05Simulates credential access and sensitive environment file reads
06Contains prompt injection phrases designed to bypass confirmation prompts
ユースケース
01Validating AST and heuristic detection rules for malicious prompt injection
02Benchmarking security scanners and automated audit tools for Claude Code skills
03Testing security rule regression suites within AI development toolkits