The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute commands using sudo to import and provision new agent teammates (sudo 5dive agent import ...). This constitutes administrative privilege escalation for the purpose of managing system-level agent services.
[REMOTE_CODE_EXECUTION]: The skill relies on npx to download and execute the @5dive/openagent package from the npm registry at runtime. While the package is provided by the skill's author, the use of npx for dynamic execution of remote code is a notable security pattern.
[EXTERNAL_DOWNLOADS]: The skill uses curl to fetch images from pollinations.ai, a third-party image generation service, to create persona avatars when a local generator is unavailable.
[PROMPT_INJECTION]: The skill processes untrusted user input (such as names, roles, and avatar descriptions) into a YAML configuration file. This file is subsequently consumed by CLI tools for rendering and provisioning. The skill includes a mitigation step by instructing the agent to validate the persona against a schema before processing to prevent malformed data or injection attempts.
Ingestion points: SKILL.md (Gather your identity section for name, role, and avatar prompts).
Boundary markers: The skill explicitly mentions that the file is checked against the v0.2 schema and malformed files are rejected before provisioning.
Capability inventory: npx package execution, sudo command execution, curl network downloads, and file system writes (mkdir, cd).
Sanitization: Schema validation is performed by the validate and import commands.

openagent