The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a provided shell script resources/pagination-loop.sh and the hubspot CLI tool. The shell script is designed with safe practices, using shell array expansion and quoted variables to handle arguments correctly without shell injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Untrusted data enters the agent context via the output of hubspot objects list or search and the pagination loop script, which output JSONL data from the HubSpot CRM.
Boundary markers: Records are structured as JSONL, providing clear boundaries for individual data objects.
Capability inventory: The skill possesses the capability to modify CRM data via objects update, delete, and upsert commands.
Sanitization: Data is transformed using jq for reshaping into specific action payloads. Although there is no explicit instruction-stripping, the dry-run mechanism and required confirmation for larger batches mitigate the risk of accidental execution of embedded instructions.

bulk-operations