The Agent Skills Directory

[SAFE]: The skill serves as a configuration guide for a remote MCP server ("https://craft-gtm.gtmhelix.com/mcp"). All identified external resources, including the GitHub repository and npm package, are owned by the verified author infrastructure ("gtmhelix.com", "shashwatgtm").
[SAFE]: The "references/operating-principles.md" file defines a robust behavioral framework that mandates the use of "web_search" and "web_fetch" for factual grounding and prevents the generation of non-English content to ensure linguistic accuracy. These instructions enhance the safety and quality of the agent's output.
[INDIRECT_PROMPT_INJECTION]: The skill's architecture involves processing untrusted data from user inputs and web sources, which is a potential vector for indirect prompt injection. However, the skill implements significant mitigations:
Ingestion points: Data enters through user-supplied market facts and content fetched via "web_search" or "web_fetch" (referenced in Rule 2 and Rule 4).
Boundary markers: Rule 7 requires an explicit "Assumption:" prefix for any inferred information, and a specific placeholder convention "[User to add: ]" is used to demarcate missing data.
Capability inventory: The skill utilizes web retrieval tools and specialized MCP tools (e.g., "pmf_scorecard", "retention_playbook"). It does not possess capabilities for local file writing, command execution, or network exfiltration.
Sanitization: Rule 4 establishes a strict 4-tier source hierarchy to prioritize authoritative ground truth and analyst-validated data over unverified third-party content.
[NO_CODE]: The skill consists entirely of instructional Markdown files. There are no executable scripts, automated package installation steps, or dynamic context injection patterns present.

helix-mcp-craft-gtm