The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided content and repository evidence to generate visual outputs. This ingestion process is susceptible to indirect prompt injection, where malicious instructions hidden in the data could affect the agent's execution or the integrity of the generated visualization.
Ingestion points: Evidence is gathered from user input and files within the repository as specified in SKILL.md.
Boundary markers: The instructions lack explicit markers or delimiters to separate untrusted evidence from the skill's operational instructions.
Capability inventory: The skill has the capability to read repository files and write HTML artifacts to the .agent-work/tasks/ directory.
Sanitization: There are no instructions for sanitizing, escaping, or validating external content before it is embedded into the output visuals.
[DYNAMIC_EXECUTION]: The skill generates standalone HTML artifacts that include inlined JavaScript and CSS based on the gathered evidence. This involves dynamic script generation which, while intended for visualization, inherits the risks of the data being processed.
Evidence: The references/html-artifacts.md file explicitly instructs the agent to include inlined JavaScript and create interactive UI elements based on the evidence provided.

show-me