The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates broad system command execution through sudo 5dive host for unit remediation, journal access, and cron management.
[REMOTE_CODE_EXECUTION]: The skill supports downloading and installing code from external sources, including GitHub repositories and arbitrary Git URLs via 5dive plugin add <owner>/<repo> and 5dive crew install <git-url>.
[DYNAMIC_EXECUTION]: The agent import command allows for the execution of 'hooks', described as arbitrary shell scripts that run automatically on agent tool events.
[PRIVILEGE_ESCALATION]: Extensive use of sudo for management tasks, such as sudo 5dive agent create, sudo 5dive account login, and sudo 5dive plugin add, providing a mechanism for agents to perform operations with root permissions.
[CREDENTIALS_UNSAFE]: The skill manages sensitive authentication material, including API keys for various LLM providers (e.g., Anthropic, OpenAI, DeepSeek), SSH keys via 5dive fleet add, and mints temporary GitHub tokens for repository operations.
[PERSISTENCE]: The skill can establish persistence through task add --recurring for scheduled tasks and 5dive proof on, which installs a daily root cron job.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external messaging channels (Telegram, Discord) and GitHub/webhook triggers, which are then passed to agents via 5dive agent send without explicit sanitization. 1. Ingestion points: Telegram/Discord messages, GitHub issue labels, and Webhook deliveries. 2. Boundary markers: None specified in instructions. 3. Capability inventory: Root command execution, file writing (wiki, secrets), and network operations (GitHub push, fleet SSH). 4. Sanitization: No evidence of filtering or escaping external content before redirection.

5dive-cli-extras