The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's preamble and various workflow steps execute local binaries and shell commands (e.g., gstack-config, gstack-update-check, git, gh, glab) to manage configuration, detect project state, and perform repository operations.
[DYNAMIC_EXECUTION]: The skill uses eval and source on the output of its own local binaries (like gstack-slug and gstack-repo-mode) to dynamically configure the shell environment and set variables based on the current workspace context.
[DATA_EXFILTRATION]: The skill includes an opt-in telemetry system that records skill usage data (outcome, duration) to a local log file and can optionally transmit this metadata to a remote server using a dedicated binary tool (gstack-telemetry-log).
[INDIRECT_PROMPT_INJECTION]: As a review tool, the skill ingests content from external sources (project plans, design docs, code files) which could contain adversarial instructions. The skill implements mitigation by providing explicit instructions to both the primary agent and the external reviewer model to ignore embedded instructions and adhere to the review framework.
[EXTERNAL_DOWNLOADS]: The skill performs update checks via a local binary and recommends the installation of the @openai/codex utility for enhanced review capabilities. It also offers to open external documentation from 'garryslist.org' for educational purposes.

plan-ceo-review