The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of local shell commands and scripts in its preamble and throughout the review phases. It executes binaries from the ~/.claude/skills/gstack/bin/ directory to handle update checks, session tracking, and telemetry logging. It also uses the open command to present external documentation to the user upon request.
[DYNAMIC_EXECUTION]: The preamble utilizes eval and source with process substitution (e.g., source <(command)) to load configuration and environment variables from the output of local helper scripts such as gstack-slug and gstack-repo-mode.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data, including plan files and repository source code, which are used as context for multiple AI review phases. It implements protective instructions for external tools like Codex to prevent unintended access to skill definitions, though the ingestion of external files remains a standard injection surface for review tools.
[DATA_EXFILTRATION]: The skill includes an optional telemetry component that logs usage data to a remote endpoint. This behavior is transparently presented to the user through interactive prompts, allowing them to opt into 'community', 'anonymous', or 'off' modes. The skill explicitly states that no source code or sensitive repository data is transmitted.

autoplan