The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute the shell command openssl rand -base64 48 in a local shell to generate random seed strings used for creative design divergence (Section 5). This is a localized, benign utility operation.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a theoretical attack surface for indirect prompt injection due to its core function of processing external marketing data.
Ingestion points: The agent is instructed to read account state, subscriber lists, and recent campaign content from third-party ESP APIs (e.g., Klaviyo, Nitrosend) to audit accounts and build flows (Section 2).
Boundary markers: The skill implements robust human-in-the-loop 'Hard gates' (Section 0) requiring an explicit 'send it' confirmation from the user, which serves as a critical verification boundary for any processed data.
Capability inventory: The agent can mutate ESP state (creating segments, flows, and campaigns) and execute shell commands for string generation.
Sanitization: The skill mandates an 'Anti-slop copy protocol' (Section 4) to lint and humanize copy and requires the use of 'Safe substrates' like MJML or React Email (Section 5) to compile email structures into inbox-safe HTML, providing validation of generated code.
[SAFE]: The skill provides explicit security guidance (Security & Safety Notes) to keep sensitive credentials, such as ESP API keys, out of prompts, templates, and logs, recommending environment-based configuration instead.
[SAFE]: The skill defines strict operational boundaries (Section 0) that prohibit autonomous mass sending, scheduling to multiple recipients without approval, or probing mutating API endpoints, ensuring user oversight for all high-risk actions.

email-marketing-bible