The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The AGENTS.md file instructs the agent to install the compose-agentsmd package globally via npm (npm install -g compose-agentsmd) if it is not present. This is an unversioned package installation from an external source, which can pose a supply chain risk.
[REMOTE_CODE_EXECUTION]: The agent-ruleset.json file configures the agent to fetch rules from a remote vendor repository (github:metyatech/agent-rules@HEAD). This allows the agent's core instructions to be updated externally at runtime.
[DYNAMIC_EXECUTION]: The skill uses npx -y @metyatech/ai-quota to check usage limits before performing actions. This represents the dynamic execution of a vendor-provided CLI tool.
[INDIRECT_PROMPT_INJECTION]: The skill's work discovery loop ingests data from external sources that can be controlled by third parties, such as GitHub issues and pull requests.
Ingestion points: SKILL.md (GitHub open issues, PR reviews, notifications, and Dependabot alerts).
Boundary markers: Absent; the instructions do not specify the use of delimiters or sanitization warnings when processing external data.
Capability inventory: The skill has significant autonomy, including spawning agents to modify code, manage repositories, and provide blanket approval for operations in user-owned repos.
Sanitization: No sanitization or validation logic is mentioned for the ingested GitHub content.
[EXTERNAL_DOWNLOADS]: The skill references and downloads external configurations (rulesets) and recommends the installation of global npm packages as part of its setup and operational flow.

autonomous-orchestrator