The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to manage and communicate across a multi-agent swarm by ingesting data from GitHub issues and a custom communication bus. This architecture creates a vulnerability surface where instructions retrieved from these sources could influence agent behavior.
Ingestion points: The GitHubCoordination class reads and creates GitHub milestones and issues, while the SwarmCommunication class processes messages through the QuicSwarmBus (SKILL.md).
Boundary markers: The instructions do not provide delimiters or instructions to the agent to ignore potentially malicious content embedded in the external data it processes.
Capability inventory: The skill executes shell-based tasks via a Task() abstraction, performs repository management via the gh tool, and uses network communication via the swarm bus.
Sanitization: No sanitization, validation, or escaping logic is described for data entering the system from the GitHub API or the inter-agent bus.
[COMMAND_EXECUTION]: The coordination logic relies on triggering shell commands and CLI tools to progress through project phases.
Evidence: Multiple instances of Task() invocations are used to trigger agent activity, and the documentation provides npm run commands for executing specific phases of the implementation (SKILL.md).

V3 Swarm Coordination