The Agent Skills Directory

[COMMAND_EXECUTION]: The project instructions in AGENTS.md require the agent to execute the compose-agentsmd CLI tool and install it via npm or npx if it is not present in the environment.
[EXTERNAL_DOWNLOADS]: The skill downloads and updates project-specific instructions (AGENTS.md) from a remote repository (github:metyatech/agent-rules) at runtime using the compose-agentsmd tool.
[INDIRECT_PROMPT_INJECTION]: The skill implements a 'Session gate' that forces the agent to synchronize its instructions from a remote source before proceeding with any task. This creates an attack surface where the agent's operating rules can be modified by remote updates to the metyatech/agent-rules repository.
Ingestion points: Instructions are fetched from github:metyatech/agent-rules and written to AGENTS.md.
Boundary markers: None present for the fetched instructions.
Capability inventory: The agent is instructed to run shell commands (npm, npx, compose-agentsmd) based on these rules.
Sanitization: No evidence of validation or sanitization of the remote instruction content.

cli-design