The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the task utility to execute diagnostic and management commands for LLM serving infrastructure. These include hardware monitoring (task gpu), process control (task serve, task restart, task stop), and performance benchmarking (task bench). These operations are consistent with the skill's purpose as an incident runbook.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface by ingesting server logs which may contain untrusted data from user prompts.
Ingestion points: System logs are read using task logs in both SKILL.md and references/symptoms.md.
Boundary markers: There are no explicit delimiters or instructions to the agent to disregard instructions embedded within the logs.
Capability inventory: The agent has the ability to execute service management commands like task serve, task restart, and task stop based on its analysis.
Sanitization: No specific filtering or sanitization of log content is described.
[EXTERNAL_DOWNLOADS]: The skill references external documentation from official sources, including vLLM, SGLang, and roadmap.sh. These links provide informational context for troubleshooting and do not initiate automated downloads or code execution.

roofline-incident