The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads and executes the Bun runtime installer from https://bun.sh/install. It implements a SHA-256 checksum verification step to ensure the integrity of the downloaded script before it is executed via the shell.
[COMMAND_EXECUTION]: Extensively uses local binaries provided by the author in ~/.claude/skills/gstack/bin/ for core functionality, including configuration management, session logging, and synchronization. It also employs dynamic execution techniques like eval and source with process substitution for state recovery.
[DATA_EXFILTRATION]: Implements a telemetry system that records skill usage metrics (e.g., duration, outcome, and session IDs). This system is designed with a focus on privacy, requiring user opt-in and providing options for anonymous or localized data storage.
[PROMPT_INJECTION]: Contains instructions that direct the agent to treat the skill as a primary execution path rather than a reference. It also creates a surface for indirect injection by ingesting untrusted data from web pages during visual audits via the browse and WebSearch tools, though no active exploitation is present.

design-review