The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's global rules in AGENTS.md instruct the agent to use sudo or the Windows runas /user:Administrator command for operations requiring elevated privileges, allowing for administrative actions on the host system.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because its core autonomous loop involves continuously reading and processing work from external, untrusted sources.\n
Ingestion points: Processes data from GitHub issues, pull request reviews, notifications, and entire repository contents.\n
Boundary markers: No explicit delimiters or isolation instructions are defined to separate untrusted data from the agent's logic.\n
Capability inventory: The orchestrator has broad capabilities including file modification, shell command execution, and sub-agent spawning.\n
Sanitization: Relies on a high-level behavioral review checklist rather than input-level sanitization.\n- [PROMPT_INJECTION]: The skill employs behavioral persistence by using rule files (AGENTS.md) as persistent memory. It is instructed to autonomously update these rules and its own behavioral instructions across sessions to fix gaps or redundancy.\n- [EXTERNAL_DOWNLOADS]: The skill downloads and installs various CLI tools and packages from the author's scope at runtime, including @metyatech/task-tracker, @metyatech/thread-inbox, and compose-agentsmd.

autonomous-orchestrator