The Agent Skills Directory

[SAFE]: The skill is a comprehensive framework for managing marketing context, industry benchmarks, and campaign execution using a 12-part methodology.
[SAFE]: Implements a strict 'Credential Isolation' model (documented in agency-operations-guide.md and crm-integration-guide.md). Secrets are stored in ~/.claude-marketing/credentials/ with restricted OS-level permissions (chmod 600) and the system enforces hard boundaries between brand accounts.
[SAFE]: Features an 'Approval Framework' (approval-framework.md) that classifies marketing actions by risk level. High-risk actions (e.g., ad budget changes, bulk email sends, CRM data deletion) require explicit user confirmation or multi-step compliance reviews.
[SAFE]: Includes a 'Content QA System' (eval-framework-guide.md) that orchestrates specialized Python scripts to score content for brand voice, readability, hallucination risk, and factual verifiability. This serves as a significant mitigation against AI-generated errors and indirect prompt injection.
[SAFE]: All external service integrations (Google Analytics, HubSpot, Salesforce, DeepL, etc.) reference well-known technology providers and follow standard OAuth or API key practices.
[SAFE]: References to external package installation (e.g., npx -y @anthropic-ai/mcp) occur within developer-focused documentation intended to guide user setup of their own Model Context Protocol (MCP) servers, rather than arbitrary runtime execution of untrusted code.

context-engine