The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the Bun runtime installer from https://bun.sh/install, which is a well-known service.
[REMOTE_CODE_EXECUTION]: Executes the downloaded Bun installer script using the system shell (bash). The process includes a SHA256 checksum verification step to ensure file integrity before execution.
[COMMAND_EXECUTION]: Utilizes extensive shell commands for environment preparation, configuration management, and execution of local binaries within the gstack tool suite (e.g., gstack-config, gstack-update-check, gstack-slug).
[DATA_EXFILTRATION]: Implements telemetry logging to track skill usage, duration, and device identifiers, which are sent to the vendor's infrastructure using the gstack-telemetry-log utility. The preamble explicitly informs the user about this collection.
[PROMPT_INJECTION]: Exposes an indirect prompt injection surface (Category 8) by processing untrusted data.
Ingestion points: Reads project-specific files such as README.md and package.json, and ingests external content via WebSearch results.
Boundary markers: The skill lacks explicit boundary markers or instructions to ignore embedded commands when processing external data.
Capability inventory: The agent has capabilities to execute Bash commands and perform Write or Edit operations on the local filesystem.
Sanitization: No sanitization or filtering of external data is performed before it is used to inform design proposals.
Mitigation: The risk is mitigated by the mandatory use of the AskUserQuestion tool, which requires human approval before the design system is finalized or written to files.

design-consultation