The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the download and execution of external code. Specifically, 5dive agent import allows importing agent packs which can contain arbitrary shell hooks that execute on tool events. While hooks are stripped by default, the --allow-hooks flag enables their execution. Additionally, 5dive crew install permits installing CrewAI workloads from arbitrary Git URLs and running them on the host.- [COMMAND_EXECUTION]: The skill relies heavily on sudo for administrative tasks such as agent creation, account management, and fleet operations. It also supports user-defined metric commands (--metric-cmd) and shell hooks, providing multiple paths for system command execution.- [CREDENTIALS_UNSAFE]: The skill handles sensitive credentials including API keys for various providers and SSH private keys (~/.ssh/id_ed25519) for fleet management. These secrets are managed via CLI flags or stored in configuration files, presenting a risk of credential exposure.- [DATA_EXFILTRATION]: The 5dive fleet command enables controlling agents on remote hosts via SSH. Furthermore, the skill's integration with Telegram and Discord channels facilitates sending data to external messaging platforms.- [EXTERNAL_DOWNLOADS]: The skill downloads agent personas from a marketplace and project code from Git repositories. Although the market is a vendor resource, the ability to download from any Git URL introduces a dependency on external content.- [PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from market packs, Git repositories, and messaging channels. While it includes boundary markers like disclosure prompts and hook stripping, its extensive capabilities—including fleet control and shell execution—could be targeted by malicious instructions embedded in processed data.

5dive-cli-extras