The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it processes setup requests and external MCP tool schemas while possessing powerful capabilities such as file system modification and command execution.
Ingestion points: Processes user setup requests and connected MCP tool schemas and annotations (SKILL.md).
Boundary markers: Instructions explicitly require the agent to wait for host/agent approvals and specific user instructions before proceeding with installation tasks (runtime.md).
Capability inventory: Includes command execution via shell and PowerShell, file writing for installation, and network access for downloading components (SKILL.md, runtime.md).
Sanitization: Implements mandatory SHA-256 checksum verification for downloaded scripts and requires the agent to inspect script contents before requesting execution approval (runtime.md).
[EXTERNAL_DOWNLOADS]: Fetches installer scripts from the vendor's official domain (get.qodo.ai). The skill requires an exact SHA-256 digest match to be verified before any file is executed, mitigating risks associated with remote downloads.
[COMMAND_EXECUTION]: Executes various CLI commands for identity verification (whoami), tool discovery (tools), and authentication (login). The instructions include warnings against bypassing security controls like the system keychain or offering unrestricted command approvals.
[REMOTE_CODE_EXECUTION]: The skill facilitates the download and execution of installation scripts (install.sh, install.ps1). This is the primary intended function of the skill and is protected by mandatory checksum verification and explicit user authorization gates.

qodo-setup