The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill executes a remote package using npx create-surf-app . to scaffold the project structure. This represents execution of code from the NPM registry.
[EXTERNAL_DOWNLOADS]: The skill performs npm install in both the backend and frontend directories, downloading numerous third-party dependencies from the NPM registry.
[COMMAND_EXECUTION]: The skill runs multiple shell commands, including starting development servers (npm run dev), inspecting local files (cat, grep), and performing local network health checks (curl to localhost).
[PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection by instructing the agent to read and obey "rules" defined in generated project files.
Ingestion points: CLAUDE.md, frontend/node_modules/@surf-ai/sdk/README.md, frontend/node_modules/@surf-ai/theme/CHARTS.md.
Boundary markers: Absent; the agent is explicitly told to read these files "BEFORE writing any code" and to follow the rules found within.
Capability inventory: Remote package execution (npx), package installation (npm), network requests (curl), and file system access (cat, grep).
Sanitization: None; the agent is directed to treat the content of these external files as authoritative project constraints.

surf-app