The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of bin/rails runner to execute arbitrary Ruby code within the application environment (e.g., in references/debugging.md and references/rails-patterns.md). This allows for full access to the application context and database. If the agent incorporates unvalidated user input into these commands, it creates a vector for arbitrary code execution.
[EXTERNAL_DOWNLOADS]: In references/sqlite-production.md, the skill provides instructions to download the Litestream binary directly from an individual's GitHub repository (github.com/benbjohnson/litestream). While the tool is known in the community, downloading binaries from personal repositories poses a supply chain risk.
[REMOTE_CODE_EXECUTION]: The skill provides a multi-step installation sequence (wget ... | tar ... | sudo mv ...) for external software in references/sqlite-production.md. This represents remote code execution of a binary from an external source.
[COMMAND_EXECUTION]: The skill uses sudo mv in references/sqlite-production.md to install a downloaded binary into a system directory (/usr/local/bin/), which is a privilege escalation pattern.
[COMMAND_EXECUTION]: In references/kamal-deployment.md, the skill uses kamal commands for server infrastructure management. This includes high-risk and potentially destructive operations such as kamal setup, kamal app stop, and kamal remove, which gives the agent significant control over production environments.
[COMMAND_EXECUTION]: The skill utilizes a custom bin/log-compressor script in references/debugging.md. The use of shell arguments like --context and --watch derived from user-provided data could lead to shell command injection if not properly sanitized by the underlying script.

rails