The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied content ('Target') to generate workflow simulations and visualizations.
Ingestion points: Input fields in the decompose, simulate-state-machine, and simulate-callstack sub-skills.
Boundary markers: The interactive-artifact-spec.md establishes clear boundaries by requiring all target-derived content to be treated as untrusted data.
Capability inventory: The skill can read repository files and write artifacts to the file system, including Mermaid diagrams and HTML files.
Sanitization: The specification requires using textContent for DOM updates and mandates encoding for the </script> sequence to prevent cross-site scripting in generated HTML.
[DYNAMIC_EXECUTION]: The simulate-state-machine skill generates a standalone HTML application that contains embedded JavaScript for rendering and interactivity.
Evidence: The skill populates the app-template.html template with machine data to create a functional visualization tool.
Mitigation: The implementation uses a declarative state machine interpreter and explicitly prohibits the use of eval() or the execution of target-derived code.
[COMMAND_EXECUTION]: The skill uses external tools to validate its outputs.
Evidence: The simulate-state-machine and simulate-callstack components attempt to use an available Mermaid CLI or library for syntax validation.

workflows