The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads and executes the official Bun installation script via the shell. Evidence: curl -fsSL "https://bun.sh/install" -o "$tmpfile" && bash "$tmpfile". The skill mitigates risk by verifying the SHA-256 checksum of the script before execution.
[COMMAND_EXECUTION]: The skill uses eval to execute shell commands produced by a local vendor-provided binary. Evidence: eval "$(~/.claude/skills/gstack/bin/gstack-slug 2>/dev/null)".
[COMMAND_EXECUTION]: The skill starts a local Python HTTP server bound to the loopback interface for live code previews. Evidence: python3 -m http.server 0 --bind 127.0.0.1.
[PROMPT_INJECTION]: The instructions contain logic to bypass user interactive prompts (AskUserQuestion) when an orchestrator session is detected, which increases agent autonomy and reduces human oversight. Evidence: If SPAWNED_SESSION is "true" ... Do NOT use AskUserQuestion for interactive prompts..
[PROMPT_INJECTION]: The skill processes project-level data (CEO plans, design metadata) without explicit boundary markers or content sanitization, creating a surface for potential indirect prompt injection. Ingestion points: approved.json, ~/.gstack/projects/$SLUG/ceo-plans/*.md.
[EXTERNAL_DOWNLOADS]: Downloads the Bun runtime installer and references the Pretext library via a public CDN. Evidence: https://bun.sh/install, https://esm.sh/@chenglou/pretext.

design-html