The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Python subprocess module to execute various system administrative tools across several files.
Evidence in SKILL.md: Use of subprocess.run to execute kubectl rollout restart for pod management.
Evidence in references/automation-toil.md: Use of subprocess.run for system inspection (df) and file cleanup (find).
Evidence in references/incident-chaos.md: Execution of kubectl delete and tc (traffic control) to simulate system failures.
[DYNAMIC_EXECUTION]: The skill implements a runbook automation framework that executes commands provided as strings through a shell environment.
Evidence in references/automation-toil.md: The AutomatedRunbook class uses subprocess.run(step.command, shell=True), which allows for the execution of arbitrary shell strings.
[PRIVILEGE_ESCALATION]: The skill requires and performs actions that involve high-level system permissions.
Evidence in references/automation-toil.md: Includes commands to restart system services via systemctl restart and delete system logs.
Evidence in references/incident-chaos.md: Includes commands to modify network configurations via iptables and tc, which typically require root or administrative access.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection due to its ingestion of external operational data combined with high-privilege capabilities.
Ingestion points: The skill is designed to ingest and analyze architecture diagrams, incident logs, and toil reports as defined in the Core Workflow of SKILL.md.
Boundary markers: There are no explicit instructions or delimiters used to prevent the agent from following instructions embedded within the ingested data.
Capability inventory: The skill has broad capabilities, including executing shell commands, managing Kubernetes resources, and modifying system/network configurations.
Sanitization: The AutomatedRunbook implementation does not demonstrate any sanitization or validation of the step.command string before it is executed via the shell.

sre-engineer