The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill includes instructions to install the 'Vite+' tool using curl -fsSL https://vite.plus | bash. This pattern involves piping a remote script from an untrusted source directly into a shell, allowing for arbitrary code execution on the user's system without prior verification.
[EXTERNAL_DOWNLOADS]: The skill promotes the installation of a third-party GitHub CLI extension gh-llm from the ShigureLab repository. This repository and its owner are not verified or listed as trusted vendors, posing a risk of executing unvetted external code.
[COMMAND_EXECUTION]: The toolkit grants the agent access to a wide array of powerful system utilities, including text manipulation (sd), network debugging (httpie), and process management (procs, btm). While useful, these capabilities can be misused if the agent is manipulated into performing unintended system actions.
[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data from GitHub PRs, issues, and logs.
Ingestion points: Commands like gh llm pr view and gh pr view fetch content from external GitHub repositories.
Boundary markers: The instructions do not provide delimiters or specific guidelines to prevent the agent from following instructions embedded within the fetched GitHub content.
Capability inventory: The agent has access to system-level commands, network tools, and file modification utilities (sd, http, curl, bash).
Sanitization: There is no evidence of sanitization or validation to ensure that data retrieved from external sources is treated as passive text rather than active instructions.

agent-cli-toolkit