The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple local binaries located in ~/.claude/skills/gstack/bin/ for various operational tasks including telemetry logging (gstack-telemetry-log), configuration management (gstack-config), and lifecycle tracking (gstack-timeline-log).
[REMOTE_CODE_EXECUTION]: Employs dynamic code execution patterns by using eval on the output of the gstack-slug tool and source on the process substitution output of gstack-repo-mode. It also uses the codex exec CLI to run architectural and strategic reviews on project code.
[PROMPT_INJECTION]: Presents an indirect prompt injection surface by reading untrusted project data (such as CLAUDE.md, TODOS.md, and plan files) and passing this content to subagents and external review CLIs. It mitigates this risk using explicit 'Filesystem Boundary' instructions for the Codex CLI to prevent it from following instructions found within skill files.
[EXTERNAL_DOWNLOADS]: Performs network-based operations including git fetch and git merge within the GBrain synchronization logic to keep session memory updated, and uses platform-specific CLIs (gh, glab) to retrieve repository and pull request metadata.

autoplan