The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill installs the Replay browser via npx replayio install and requires the @replayio/playwright package. These are external dependencies not included in the trusted source list.
REMOTE_CODE_EXECUTION (MEDIUM): The skill instructs the user to configure a remote MCP server located at https://dispatch.replay.io/nut/mcp. This allows an external service to influence agent behavior through tool definitions.
COMMAND_EXECUTION (LOW): Multiple commands are executed via Bash tools (npm, npx, yarn, pnpm, bun) to install software and run tests.
DATA_EXFILTRATION (LOW): The REPLAY_API_KEY is passed in an Authorization header to the Replay dispatch URL. While this is necessary for the service's functionality, it involves the transmission of sensitive credentials to a non-whitelisted domain.
INDIRECT_PROMPT_INJECTION (LOW):
Ingestion points: The agent receives data from the mcp__replay tool output.
Boundary markers: No specific delimiters or safety instructions are provided to the agent for processing the remote tool's output.
Capability inventory: The skill has broad execution capabilities through the permitted Bash tool variants.
Sanitization: There is no explicit sanitization logic for data returned from the MCP server.

replay-playwright