The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to invoke emacsclient, which executes Elisp code that subsequently runs shell commands (swiftc and a compiled binary) to perform the rendering.
[REMOTE_CODE_EXECUTION]: The skill generates, compiles, and executes Swift programs at runtime based on untrusted data derived from user context (prose, diagrams, or existing code). This dynamic execution of code influenced by external prompts poses a risk of executing arbitrary logic on the host system if the agent converts malicious intent into valid Swift code.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes unvalidated data from the interaction context to construct executable logic.
Ingestion points: Interaction context (processed in SKILL.md to derive views).
Boundary markers: Absent; the skill does not use delimiters to isolate user-provided data from the Swift code template.
Capability inventory: File system writes (with-temp-file), shell command execution (call-process-shell-command), and execution of arbitrary compiled binaries.
Sanitization: Absent; the user-derived content is inserted into the Swift template without validation or escaping beyond basic shell quoting for file paths.

swiftui-preview