The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits a significant attack surface for indirect prompt injection as it is designed to ingest and process data from external URLs (via mktg init --from <url> and firecrawl) and social media signals (via last30days). While the skill mandates multiple quality gates—including an editorial pass, an AI slop audit, and content reviewer agents—these are primarily focused on output quality rather than adversarial input sanitization. The reliance on external data to populate core brand files could allow malicious instructions in processed websites to influence downstream agent behavior.
[COMMAND_EXECUTION]: The skill relies extensively on the Bash tool to execute mktg CLI commands, curl for interacting with a local Studio API, and jq for data processing. These operations are core to its function as a marketing orchestrator and are used for local state management, health checks, and dashboard integration.
[SAFE]: The skill demonstrates high-security awareness by explicitly instructing the agent to never read or write .env files and to avoid hardcoding credentials, instead referencing a stack.md configuration. Furthermore, the rules/safety.md file mandates the use of --dry-run and explicit user approval before any external action, such as sending emails or posting to social media, which provides a strong human-in-the-loop safety mechanism.

cmo