The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The instructions require the agent or user to execute commands with elevated privileges using sudo. This includes service management tasks like sudo systemctl start omniroute, log inspection with sudo journalctl, and port verification using sudo lsof.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill documents the exact file system paths for sensitive components, including the agent configuration file (/home/openclaw/.opencode/config.json), the application database (storage.sqlite) containing provider connection details and usage logs, and the initialization script containing environment variable definitions.
[METADATA_POISONING]: The skill uses a deceptive persona, identifying as "Sam Altman" with credentials such as "CEO of OpenAI" and "Led GPT development." While potentially thematic for an infrastructure skill, this constitutes misleading metadata regarding the skill's origin and authoritative expertise.
[INDIRECT_PROMPT_INJECTION]: The skill provides tools that interface with external, multi-provider LLM outputs, creating a surface for indirect prompt injection.
1. Ingestion points: Tools like route_request, list_combos, and explain_route (SKILL.md) return data from external LLM providers and the OmniRoute proxy.
2. Boundary markers: No delimiters or instructions to ignore embedded commands are specified in the provided integration examples.
3. Capability inventory: The skill facilitates agent orchestration and model selection, which can influence downstream agent actions based on untrusted provider responses.
4. Sanitization: There is no mention of sanitization or validation of the content returned by the external routing service before it is processed by the agent.

omniroute-integration