The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a 'Remote Execution' helper that uses the salt cmd.run command, enabling the agent to execute arbitrary shell commands on any Salt minion. This is a powerful administrative capability that allows for high-impact changes across the infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from remote minions, such as grain data, pillar data, and command execution results, which could contain malicious instructions designed to influence the agent's logic.
Ingestion points: Output from salt minion commands, grain inspection, pillar retrieval, and salt-api responses.
Boundary markers: Not present. The skill does not implement delimiters or safety instructions to distinguish between retrieved data and its own administrative logic.
Capability inventory: Arbitrary command execution on minions (salt cmd.run), state application (state.apply), and access to configuration files (cat /etc/salt/master.d/reactor.conf).
Sanitization: While the skill uses jq to parse structured JSON output, it does not sanitize or escape the content within those fields to prevent instruction injection.

managing-saltstack