The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill takes a project name as a positional argument ($1) which is interpolated into shell commands for project generation. This creates a potential surface where a malicious user could provide a crafted project name to attempt command injection. However, the skill explicitly instructs the agent to 'Validate the project name and destination' and 'refuse an existing/symlinked target' before scaffolding.
Ingestion points: SKILL.md uses the argument-hint [project name] and references it as $1 in the instruction text.
Boundary markers: No explicit delimiters (like triple quotes) are defined for the interpolation in the instructions.
Capability inventory: The skill has access to rails, bundle, npm, yarn, and bin/* via allowed-tools.
Sanitization: The skill contains instructions to validate the project name and destination before execution.
[COMMAND_EXECUTION]: The skill facilitates the execution of local development tools including rails, bundle, npm, and yarn. These are scoped to specific subcommands in the allowed-tools frontmatter, limiting the potential for arbitrary command execution.
[SAFE]: The skill recommends several security best practices for the generated application, such as using the brakeman gem for static security analysis and implementing encrypted fields for sensitive data like OAuth tokens and API keys.

new-rails-project