The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions and generated code templates utilize permissionMode: "dangerous". This configuration instructs agent environments (such as Claude Code) to bypass standard security prompts (mapping to --permission-mode bypassPermissions), enabling the agent to perform sensitive actions like file system modifications and shell command execution without user confirmation.
[EXTERNAL_DOWNLOADS]: The skill workflow requires the agent to fetch documentation from the vendor's domain https://docs.dynobox.xyz/config-authoring.md using curl. Ingesting external content directly into the agent's context during skill execution introduces a risk of indirect instruction injection if the source is manipulated.
[COMMAND_EXECUTION]: The skill performs runtime validation of generated code by executing node --check and npx dynobox validate on files it has just written to the file system, involving the execution of code derived from external input.
[PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface by processing untrusted SKILL.md files to generate test scenarios. Ingestion points: Reads SKILL.md files from the filesystem; Boundary markers: None identified to separate skill content from generator instructions; Capability inventory: Includes mkdir, cp, node, and npx; Sanitization: No evidence of input validation or escaping before content is interpolated into generated prompts or assertions.

dyno-from-skill