The Agent Skills Directory

[DYNAMIC_EXECUTION]: The code in EXAMPLES.md uses the .constantize method in Rails to dynamically resolve class names based on strings extracted from the Accept request header. While a regex is used for basic filtering, this pattern of dynamic dispatch to classes derived from external input is risky and specifically prohibited by the skill's own HARD-GATE safety instructions in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing untrusted data which could lead to logic bypass or injection.
Ingestion points: Untrusted data enters the context via request.headers['Accept'] and request.path (as seen in EXAMPLES.md).
Boundary markers: The SKILL.md includes a HARD-GATE requiring explicit allowlists for version resolution, but the actual implementation examples lack these markers.
Capability inventory: The skill is designed to modify core application files (controllers, routes) and executes shell commands via bundle exec rspec.
Sanitization: Only basic regex filtering (\\d+) is present in the examples, which does not safely prevent the dynamic resolution of unintended constants when using .constantize.
[COMMAND_EXECUTION]: The skill instructions and workflow files (SKILL.md, references/workflow.md) explicitly require the agent to execute shell commands (bundle exec rspec) for regression testing. This represents a capability for local command execution within the agent's environment.

version-api