The Agent Skills Directory

[COMMAND_EXECUTION]: The skill explicitly instructs the agent to use sudo for tasks that require elevated privileges in AGENTS.md under the 'Workflow and command execution' section.- [COMMAND_EXECUTION]: Implements a 'blanket approval' policy for user-owned repositories, allowing the agent to approve its own plans and perform state-changing operations without explicit human confirmation.- [EXTERNAL_DOWNLOADS]: The skill requires the installation and execution of several external tools from the metyatech namespace, including @metyatech/task-tracker, @metyatech/thread-inbox, @metyatech/ai-quota, and compose-agentsmd.- [EXTERNAL_DOWNLOADS]: The skill's ruleset configuration in agent-ruleset.json targets a remote repository github:metyatech/agent-rules@HEAD for dynamic rule composition.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it autonomously scans untrusted external sources like GitHub issues, PR comments, and package registries during its discovery loop.
Ingestion points: GitHub issues, PR reviews, notifications, and dependabot alerts (identified in SKILL.md).
Boundary markers: None identified in the work discovery or processing logic.
Capability inventory: Extensive capabilities including file modification, sudo command execution, and multi-agent delegation (identified in SKILL.md and AGENTS.md).
Sanitization: No specific sanitization or escaping procedures are defined for data ingested from external sources.

autonomous-orchestrator