The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several local utilities located in ~/.claude/skills/gstack/bin/ for session management, telemetry, and project context recovery. These include gstack-skill-start, gstack-slug, and gstack-learnings-log. These tools are part of the vendor's infrastructure for the gstack system.
[INDIRECT_PROMPT_INJECTION]: The skill reads external configuration files such as fly.toml, render.yaml, and GitHub Actions workflow files to detect deployment platforms. While these files are local to the user's project, their content influences the agent's behavior. The skill mitigates risks by using AskUserQuestion to confirm all inferred settings with the user before persisting them to CLAUDE.md.
[DATA_EXFILTRATION]: The skill uses curl to perform health checks against inferred or user-provided production URLs. It includes explicit instructions to redact sensitive information, such as showing only the first few characters of environment variables like RENDER_API_KEY and mandating that captured secrets never appear in logs or chat output.
[EXTERNAL_DOWNLOADS]: The skill mentions the Aside browser (aside.com) as a recommended tool for handling third-party web actions. It performs a local probe to check for the tool's presence but does not attempt to download or install it automatically, relying instead on user-initiated setup.

setup-deploy