The Agent Skills Directory

[OBFUSCATION]: The file example.html uses HTML character entities to encode specific command strings within a code display section.
Evidence: The string rm -rf / decodes to rm -rf / and drop table decodes to drop table.
Context: These are used in a ts-codebox within the forbidden_always section of a safety policy slide to illustrate blocked actions. While technically obfuscated, the context indicates these are non-executable examples for display.
[INDIRECT_PROMPT_INJECTION]: The skill provides an HTML/CSS template intended to be populated with external data, creating an ingestion surface for potentially malicious instructions.
Ingestion points: The slide content areas in example.html (e.g., .ts-sub, .ts-alert-box, .ts-check .txt).
Boundary markers: No delimiters or instructions are present to prevent the agent from obeying instructions embedded in the injected text.
Capability inventory: The skill is restricted to visual presentation; it has no file-write, network, or code execution capabilities.
Sanitization: No sanitization logic for external input is included in the static files.
[EXTERNAL_DOWNLOADS]: The skill references external assets from a well-known service.
Evidence: example.html imports multiple fonts from fonts.googleapis.com via CSS @import rules.

deck-safety-alert