The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains multiple prompt injection patterns such as 'Ignore all previous instructions' and 'DAN mode'. These are located in references/prompt-injection-patterns.md and are clearly documented as threat patterns for the scanner to detect in other skills, rather than instructions for the agent to follow.
[MALICIOUS_CODE]: The file references/dangerous-code-patterns.md lists various malicious code snippets including reverse shells, credential theft, and data exfiltration. These are explicitly provided as examples of what to audit in other skills and do not represent executable payloads for the skill itself.
[DYNAMIC_EXECUTION]: The bundled script scripts/scan_skill.py uses regex to identify dangerous dynamic execution patterns (like eval() or exec()) in target files. The script itself uses standard libraries and performs safe file operations.
[OBFUSCATION]: The skill includes detection logic for Unicode steganography and zero-width characters. The presence of these patterns in the scripts and references is for detection purposes and does not represent an attempt to hide malicious intent in the skill itself.
[INDIRECT_PROMPT_INJECTION]: As a security scanner, the skill is designed to ingest and report on potentially malicious content from other skills. The instructions include specific guidance for the agent to evaluate the context and intent of findings to prevent being misled by the data it is auditing.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute its bundled Python script using the uv manager. This use is justified by the tool's primary function of performing automated security scans and is properly scoped in the instructions.

skill-scanner