The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple command-line utilities to manage the development stack lifecycle.
Evidence: scripts/boot.sh invokes portless, emulate, agent-browser, git, tailscale, and jq to configure the environment and start services.
Evidence: scripts/stop.sh uses pgrep and kill to perform a recursive process tree teardown in reverse boot order.
[INDIRECT_PROMPT_INJECTION]: The skill ingests project metadata to customize boot behavior and surface developer hints, creating a limited indirect injection surface.
Ingestion points: package.json, turbo.json, and pnpm-workspace.yaml are read in scripts/boot.sh to detect dependencies and monorepo configurations.
Boundary markers: Absent.
Capability inventory: Subprocess execution of development tools, local state file writes in .claude/state/, and network configuration via Tailscale.
Sanitization: Employs jq for structured data extraction and tr for branch-to-slug normalization to ensure data is handled safely.
[EXTERNAL_DOWNLOADS]: The skill references and uses external software packages to facilitate development workflows.
Evidence: scripts/boot.sh uses npx --yes wait-on to perform network-based health checks on the dev server.
Evidence: Documentation recommends global installation of portless, emulate, and agent-browser (tools from the Vercel Labs ecosystem).

dev