The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by rendering live web pages and SERP results to perform SEO audits. This creates a surface for indirect prompt injection if the ingested content contains adversarial instructions.
Ingestion points: In SKILL.md (Workflow steps 2, 3, 4), the agent is instructed to use browser/html, browser/markdown, and seo/serp to fetch external content.
Boundary markers: The skill includes a "Guardrails" section explicitly stating a "read-only" posture ("eyes, not hands") and requires the agent to cite evidence from live reads.
Capability inventory: The skill is primarily designed for information gathering and reporting; it lacks high-privilege capabilities like arbitrary shell execution or local file writing.
Sanitization: No explicit sanitization or filtering of the ingested HTML/Markdown content is mentioned before processing.
[EXTERNAL_DOWNLOADS]: The README.md provides installation instructions using npx skills add unifapi-agent/agents and a plugin marketplace. These are standard installation procedures for the UnifAPI platform and are documented as connecting to the vendor's official MCP server infrastructure at https://mcp.unifapi.com over OAuth.
[COMMAND_EXECUTION]: While the skill mentions using tools like curl in its Pillar checklists, these are used as descriptions of how data is fetched via the provided unifapi tools, rather than executing arbitrary user-supplied shell commands.

seo-audit