The Agent Skills Directory

[PROMPT_INJECTION]: The references/prompt-injection-patterns.md file contains a detailed list of jailbreak attempts, role reassignment, and instruction overrides such as 'Ignore all previous instructions'. As this skill is a security scanner, these patterns serve as reference data for analysis and are not intended to be executed as instructions by the agent.
[OBFUSCATION]: The script scripts/scan_skill.py contains regex patterns designed to detect zero-width characters (e.g., U+200B), RTL overrides, and homoglyphs. These patterns, alongside documentation in the references directory, are used to identify obfuscation techniques in other skills and are not used to hide malicious behavior within this skill itself.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute its internal Python-based scanner using uv run scripts/scan_skill.py. This execution is scoped to the skill's own directory and is necessary for providing structured security findings to the user.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted files from external skill directories, creating a surface for indirect prompt injection. The skill mitigates this by instructing the agent to evaluate scanner findings as leads for deeper analysis and explicitly distinguishing between attack patterns and documentation context.

skill-scanner