The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the agent to download and install the omniroute CLI from a repository at github.com/diegosouzapw/OmniRoute. This repository is not a recognized trusted source, posing a risk of unverifiable third-party code execution.
[CREDENTIALS_UNSAFE]: The skill documentation includes the command omniroute keys reveal <id>, which allows for the extraction of full API key values. While this is a functional feature of the documented CLI, it provides a direct mechanism for an agent to expose sensitive credentials if misdirected.
[COMMAND_EXECUTION]: The skill is entirely composed of shell commands that interact with system configuration, network connections, and security credentials via the omniroute binary.
[INDIRECT_PROMPT_INJECTION]: The omniroute combo suggest --task "<task>" command provides an attack surface where untrusted data (task descriptions) could influence the agent's behavior or tool output.
Ingestion points: Untrusted text provided to the --task flag in SKILL.md.
Boundary markers: No explicit delimiters or instructions are provided to the agent to treat the task text as data only.
Capability inventory: The skill utilizes subprocess execution of the omniroute CLI and network-connected API interactions.
Sanitization: No sanitization or validation logic is specified for the task input.

omniroute-cli-providers