The Agent Skills Directory

[SAFE]: The skill serves as a structural guide and instructional prompt for building HTML mockups. Its primary functions involve reading context (briefs, assets) and generating static HTML/CSS artifacts, which is a benign use case.
[COMMAND_EXECUTION]: The SETUP.md file references the use of npx zapier-sdk login. This is a standard initialization command for the Zapier SDK, a legitimate developer tool provided by the vendor, and is intended for user-initiated setup rather than automated malicious execution.
[EXTERNAL_DOWNLOADS]: The skill references mcp.zapier.com and the zapier-sdk package. These are official, well-known resources belonging to the vendor (Zapier) and are used for legitimate integration purposes.
[DATA_EXFILTRATION]: While the skill interacts with external systems (Google Docs, CRM, Asset Libraries) via Zapier MCP/SDK, it emphasizes human-in-the-loop controls. Instructions in README.md and ONBOARDING-PROMPT.md specifically mandate explicit user approval before any external writes or message sending, mitigating the risk of unauthorized data movement.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources (e.g., page_brief, assets).
Ingestion points: SKILL.md and SCHEMA-MAP.md identify external fields like page_brief and audience as input sources.
Boundary markers: No specific technical boundary markers (like XML tags or delimiters) are defined in the instructions to isolate these inputs.
Capability inventory: The skill can log artifacts and route reviews via Zapier tools.
Sanitization: The skill relies on manual 'human approval' and 'sanitized test runs' rather than programmatic sanitization. Given the intended use as a mockup tool, this risk is assessed as low.

html-mockup-builder