The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a complex preamble and multiple setup steps that execute local binaries (e.g., gstack-config, gstack-slug, gstack-brain-sync) and shell commands to manage environment state, session tracking, and project metadata.
[COMMAND_EXECUTION]: The skill employs dynamic execution patterns, specifically eval and source <(...), to load environment variables and configurations generated by its own utility binaries.
[DATA_EXFILTRATION]: The skill includes an opt-in telemetry system and an artifact synchronization feature (gbrain). Usage data is logged locally and may be sent to remote endpoints, while project artifacts can be synced to a remote repository; both behaviors are governed by user prompts and configuration settings.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it reads untrusted project files—such as CLAUDE.md, fly.toml, render.yaml, and GitHub Action workflows—to extract deployment parameters.
Ingestion points: Reads CLAUDE.md and various platform configuration/workflow files (SKILL.md).
Boundary markers: None identified.
Capability inventory: The skill has Bash, Write, Edit, and network access (curl) capabilities.
Sanitization: No explicit sanitization or validation of extracted configuration values is mentioned before they are written back to CLAUDE.md or used in prompts.
[COMMAND_EXECUTION]: Deployment verification steps involve executing platform-specific CLI tools (e.g., fly status, vercel ls) and performing network checks via curl to confirm endpoint availability.

setup-deploy