The Agent Skills Directory

[PROMPT_INJECTION]: The file examples/prompt-injection.txt contains literal prompt injection strings, such as "Ignore all previous instructions" and "Ignore your system instructions". These are provided as educational examples of direct and indirect attacks but could be misinterpreted by the model if processed as instructions.
[CREDENTIALS_UNSAFE]: Multiple files contain hardcoded dummy credentials and API keys used to illustrate security failures. Examples include sk-abc123xyz789 in SKILL.md and examples/k8s-rbac.yaml, and database_password: "MySecurePassword123!" in examples/k8s-rbac.yaml. These are explicitly labeled as vulnerable or insecure patterns.
[EXTERNAL_DOWNLOADS]: The skill's README.md and skill.json reference an external GitHub repository (github.com/mfkocalar/OWASP-Security-Skills) for installation and updates. It also references third-party scripts like DOMPurify from cdn.jsdelivr.net in examples/xss.html.
[COMMAND_EXECUTION]: The guidance in SKILL.md and various files in the examples/ directory documents the use of potentially dangerous system-level operations, such as os.system(), subprocess.run(), and privileged container configurations, to demonstrate insecure coding practices.
[PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted code, configurations, or descriptions provided by the user. This creates an indirect prompt injection attack surface where malicious instructions could be embedded in the data being audited (Category 8).
Ingestion points: User-provided code snippets, API endpoint descriptions, and Kubernetes manifests analyzed via security prompts.
Boundary markers: The instructions in owasp-css-instructions.md do not specify required delimiters or sanitization for the input data.
Capability inventory: The skill primarily performs text analysis and reference lookups; it does not contain scripts for executing the provided code.
Sanitization: No explicit sanitization or filtering of the input code is implemented before processing.

owasp-security