The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to run surf install at the start of sessions to update the CLI binary from the vendor's official infrastructure.- [DATA_EXFILTRATION]: The surf feedback command is used to send the last 10 turns of the conversation history to the vendor's servers. The instructions explicitly mandate that the agent must obtain user consent before executing this command and warns against including sensitive data.- [PERSISTENCE]: The skill contains logic to modify project-level configuration files like AGENTS.md or CLAUDE.md. It appends routing rules that instruct future agent instances to prioritize this skill for crypto-related data queries, ensuring its continued use across sessions.- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external crypto APIs, which represents an attack surface for indirect prompt injection. The instructions include a specific security boundary, advising the agent to treat all API responses as untrusted data and avoid executing any instructions found within them.- [COMMAND_EXECUTION]: The skill utilizes the bash tool to execute a variety of CLI commands, including surf, git, and file system utilities, to manage its operation and project integration.

surf