The Agent Skills Directory

[COMMAND_EXECUTION]: Script templates utilize the subprocess module to interact with system tools and cluster management utilities.\n
Evidence: SKILL.md and references/automation-toil.md contain Python scripts that execute commands such as kubectl, systemctl, df, find, and curl to manage service states and resources.\n- [PRIVILEGE_ESCALATION]: The skill provides patterns for operations that typically require elevated system or cluster permissions.\n
Evidence: references/incident-chaos.md defines classes for manipulating kernel-level networking via traffic control (tc) and firewall rules (iptables), while references/automation-toil.md documents service restarts via systemctl.\n- [INDIRECT_PROMPT_INJECTION]: The skill documents an attack surface where external inputs flow into high-privilege commands.\n
Ingestion points: The auto-remediation template in SKILL.md processes service and namespace identifiers from command-line arguments (sys.argv).\n
Boundary markers: The code templates lack explicit delimiters or instructions to treat ingested data as untrusted.\n
Capability inventory: The skill possesses capabilities for file system deletion, network manipulation, and cluster modification.\n
Sanitization: Input variables are interpolated directly into shell command strings and PromQL queries without validation, creating a potential injection surface.\n- [DYNAMIC_EXECUTION]: A runbook automation framework within the skill allows for the execution of arbitrary command strings provided as text.\n
Evidence: references/automation-toil.md defines an AutomatedRunbook class that uses subprocess.run with shell=True to execute steps defined in runbook configuration.

sre-engineer