The Agent Skills Directory

[COMMAND_EXECUTION]: The skill extensively executes local binary tools and scripts (e.g., gstack-config, gstack-slug, gstack-repo-mode) for environment setup, configuration, and state management.
[COMMAND_EXECUTION]: Employs dynamic execution via eval and source <(...) to run shell code generated by its internal helper scripts.
[DATA_EXFILTRATION]: Collects and transmits usage analytics (skill name, duration, repository name) to vendor-owned telemetry scripts. This behavior is subject to user consent and configurable telemetry levels.
[PROMPT_INJECTION]: Implements an indirect prompt injection surface by reading project files (like DESIGN.md) and user-defined directories to construct prompts for AI subagents. It lacks explicit boundary markers or sanitization for this ingested data.
[COMMAND_EXECUTION]: Uses the Agent tool to spawn subagents for parallel design variant generation, providing them with instructions to execute the design binary.
[COMMAND_EXECUTION]: Modifies project metadata by appending skill routing rules to the CLAUDE.md file, which influences the agent's behavior in subsequent sessions.

design-shotgun