The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the irasutoya-cli package via gem install if it is not already present on the system. This involves downloading and installing third-party code from RubyGems, a public registry, which is not verified for security within the skill's context.
[COMMAND_EXECUTION]: The skill's primary functionality is driven by executing shell commands, including irasutoya search, irasutoya random, and a shell loop using grep to filter results. The use of shell-based automation increases the risk of unintended system interactions.
[INDIRECT_PROMPT_INJECTION]: The skill translates natural language user prompts into keywords for shell commands, creating a potential injection surface. While the instructions recommend quoting queries, a sophisticated adversarial prompt could attempt to escape the shell context or inject malicious sub-commands into the search queries.
Ingestion points: User-provided descriptions of images (found in SKILL.md and evals/evals.json).
Boundary markers: The instructions specify wrapping queries in double quotes (irasutoya search "query"), but do not provide formal sanitization rules for the generated Japanese keywords.
Capability inventory: Execution of the irasutoya CLI tool and installation of Ruby gems (gem install).
Sanitization: Relies on the agent to perform translation and optimization of queries into short Japanese noun phrases before execution.

irasutoya-search