The Agent Skills Directory

[SAFE]: The skill utilizes the vendor's own CLI tool (hubspot) to perform CRM data operations. The use of temporary files in /tmp for data aggregation is a standard pattern for CLI-based agent skills.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by ingesting and processing untrusted CRM message content and thread subjects.
Ingestion points: Conversation thread metadata and message text are retrieved via the hubspot conversations command group in SKILL.md (Steps 2, 4, and 5).
Boundary markers: The summary output in Step 5 and Step 6 does not utilize specific boundary markers or instructions to the agent to ignore embedded commands in the message text.
Capability inventory: The skill is capable of executing shell commands through the hubspot CLI and writing to the local filesystem via shell redirection.
Sanitization: The skill uses jq for all data extraction and formatting. This provides robust protection against JSON structural injection, though it does not filter the semantic content of the retrieved messages. This is considered acceptable as it aligns with the skill's primary purpose of data retrieval.

conversation-history-brief