The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses eval and source commands to execute output from local binaries, such as gstack-slug and gstack-repo-mode, located in the ~/.claude/skills/gstack/bin/ directory to configure the environment.
[COMMAND_EXECUTION]: The skill frequently invokes a suite of local vendor-supplied utilities (e.g., gstack-config, gstack-update-check, gstack-timeline-log) to manage settings, check for updates, and log execution milestones.
[DATA_EXFILTRATION]: The skill performs telemetry logging by sending usage data such as skill names, duration, and outcomes to the vendor's logging service via the gstack-telemetry-log binary. This behavior is transparent and can be disabled via user configuration.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided feedback and preferences from local JSON files (feedback.json, approved.json) and incorporates this data into prompts for iterative design generation and sub-agent instructions.
Ingestion points: Reads ~/.gstack/projects/$SLUG/designs/*/feedback.json and approved.json files.
Boundary markers: Missing; feedback is interpolated into prompts as natural language.
Capability inventory: Uses Bash, Agent (for parallel variant generation), and AskUserQuestion tools.
Sanitization: No explicit sanitization of feedback content is performed before interpolation.

design-shotgun