The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes contact records from HubSpot, creating a surface for indirect prompt injection. Malicious instructions embedded in CRM fields could potentially influence the agent's behavior during enrichment or reporting. Ingestion points: Contact properties including firstname, lastname, company, and jobtitle are ingested via the search_crm_objects tool as described in Stage 1 of the workflow. Boundary markers: The instructions do not define boundary markers or clear separation between the ingested data and the agent's instructions. Capability inventory: The skill uses tools to write to the CRM (manage_crm_objects), call external enrichment APIs (Apollo, Clay), and generate local HTML and JSON files. Sanitization: The workflow does not specify any sanitization or validation of the input data before it is used in tool parameters.
[DATA_EXFILTRATION]: The skill transmits sensitive contact information (PII) to external services (Apollo and Clay) for enrichment. While this is the intended behavior for this automation, it constitutes data transmission to non-whitelisted domains. Evidence: Tool calls such as apollo_people_bulk_match and find-and-enrich-contacts-at-company send data to external infrastructure.

prospect-enrich