The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the subprocess.py module to execute system-level commands.
Evidence in SKILL.md: Executes kubectl rollout restart based on external metrics.
Evidence in references/automation-toil.md: Uses subprocess.run for df, find, curl, and systemctl restart operations.
[DYNAMIC_EXECUTION]: The AutomatedRunbook class in references/automation-toil.md implements a pattern where shell commands are stored as strings and executed using subprocess.run(shell=True). This is a high-risk implementation that can lead to command injection if the input strings are not strictly controlled.
[PRIVILEGE_ESCALATION]: The skill provides templates and scripts for performing operations that require administrative or root privileges.
Evidence in references/incident-chaos.md: Includes logic for tc (traffic control) and iptables manipulation via kubectl exec, which are high-privilege operations used for chaos engineering experiments.
[EXTERNAL_DOWNLOADS]: The skill metadata and documentation reference jeffallan.github.io, which has been flagged as blacklisted by automated reputation scanners. Additionally, the skill file SKILL.md itself has been identified as potentially malicious by file reputation scanners.
[INDIRECT_PROMPT_INJECTION]: A vulnerability surface exists where data ingested from external monitoring systems triggers automated remediation actions.
Ingestion points: SKILL.md fetches JSON data from a Prometheus API endpoint using urllib.request.
Boundary markers: No explicit boundary markers or 'ignore' instructions are present to protect against malicious data returned from the API.
Capability inventory: The skill possesses the ability to perform container restarts (kubectl) and service management (systemctl) across multiple scripts.
Sanitization: The scripts lack validation or sanitization of the metrics returned by the API before they are used to trigger subprocess executions.

sre-engineer