The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches an installation script from the author's official GitHub repository (avivsinai/agent-message-queue).
[REMOTE_CODE_EXECUTION]: Executes the downloaded installation script via a shell pipe (curl | bash) as part of the prerequisite setup.
[PROMPT_INJECTION]: Instructions in the 'Quick Start' section encourage agents to use flags such as --dangerously-skip-permissions and --dangerously-bypass-approvals-and-sandbox, which are designed to override tool-internal security constraints.
[PROMPT_INJECTION]: The skill processes external data (inter-agent messages) which introduces a surface for indirect prompt injection.
Ingestion points: Messages are read from the mailbox directory via amq drain, amq list, and amq read (SKILL.md, references/coop-mode.md).
Boundary markers: The CLI implements strict header validation and moves malformed messages to a Dead Letter Queue (DLQ) (references/coop-mode.md).
Capability inventory: Capabilities include executing amq commands, modifying project files like WORKFLOW.md via symphony init, and running shell-based statusline scripts (SKILL.md, references/integrations.md).
Sanitization: Relies on the underlying CLI's header validation and RFC3339 timestamp checks (references/message-format.md).
[COMMAND_EXECUTION]: Provides shell script snippets for terminal integration, including statusline updates and tab title manipulation using ANSI escape sequences.
[COMMAND_EXECUTION]: Includes the amq integration symphony init command which modifies local project files (WORKFLOW.md).

amq-cli