The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to interact with several local command-line interfaces (CLIs) such as aws, gcloud, doctl, flarectl, and vercel to manage DNS records.
Evidence: The workflow includes steps to detect and use these CLIs for creating resource records (CNAME, TXT, etc.) on behalf of the user.
Safety Measures: The skill instructions include explicit rules to never ask for or handle registrar credentials, never install new CLIs, never echo tokens, and always show exact commands for user confirmation before execution.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied domain strings and integrates with external CLI tool outputs, which represents a potential injection surface.
Ingestion points: User-provided domain names and Convex deployment URLs in SKILL.md.
Boundary markers: The instructions do not use specific XML/markdown delimiters for user input, but rely on human oversight.
Capability inventory: Execution of powerful DNS management commands via subprocesses.
Sanitization: The skill mitigates risk by enforcing a 'confirm-before-run' policy where the agent must display the exact generated command and wait for explicit user approval before performing any modification.

convex-domains