The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a framework for managing deterministic shell jobs. The documentation explicitly states that enabling this functionality authorizes arbitrary command execution on the worker environment, which is described as a remote code execution surface.
[PERSISTENCE]: The skill defines a durable execution doctrine to ensure background tasks survive restarts. This includes the implementation of a deadman pattern which involves registering recurring host-cron checks or crontab entries to monitor job completion and report results across sessions.
[DYNAMIC_EXECUTION]: The skill facilitates the spawning of subagents and allows replaying existing jobs with modified parameters via the replay_job tool and data_overrides field, allowing for runtime changes to executed logic.
[INDIRECT_PROMPT_INJECTION]: The skill architecture is susceptible to indirect prompt injection through several ingestion points.
Ingestion points: The skill processes job parameters, fan-out manifests, and side-channel inbox messages delivered to running agents via send_job_message.
Boundary markers: The skill lacks documented boundary markers to separate external directives from the agent's core instructions.
Capability inventory: The skill manages arbitrary shell command execution and the spawning of autonomous subagents with extensive tool access.
Sanitization: There is no specific sanitization or validation logic detailed for the content of job directives or manifests.

minion-orchestrator