The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes live public data retrieved from external AI search engines, which represents a potential surface for indirect prompt injection if those sources contain adversarial instructions.
Ingestion points: Data is ingested through the unifapi skill endpoints such as geo/mentions/search and geo/serp.
Boundary markers: The skill does not explicitly define custom delimiters for tool outputs in its instructions.
Capability inventory: The skill is primarily focused on data aggregation, calculation of metrics, and report generation; it does not request high-privilege system access or command execution.
Sanitization: No specific sanitization logic for the ingested search data is mentioned in the prompt instructions.
[EXTERNAL_DOWNLOADS]: The skill requires a connection to an external MCP server for data retrieval.
Details: It connects to https://mcp.unifapi.com via OAuth.
Context: This is a vendor-owned resource provided by unifapi-agent for the skill's primary functionality.
[PERSISTENCE]: The skill reads from and updates local files to maintain configuration and track historical trends.
Details: It accesses .agents/product-marketing.md and .claude/product-marketing.md to store brand, domain, and competitor sets.
Purpose: This is operational persistence required for longitudinal trend analysis (Share-of-Voice) between runs.

llm-mention-tracking