The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data, such as HubSpot deal notes and Clari call summaries. This data is processed to generate momentum scores and re-engagement drafts, presenting a surface for indirect prompt injection attacks where a third party (e.g., a call participant or someone with CRM access) could embed instructions to manipulate the agent's output.
Ingestion points: reads from notes_last_updated, clari_get_call_summary, and general CRM activity logs.
Boundary markers: No explicit delimiters are specified for separating untrusted external content from the skill's system instructions during processing.
Capability inventory: The skill is capable of creating Gmail drafts (create_draft) and scheduling calendar events (create_event).
Sanitization: The risk is mitigated by the mandatory check_my_copy brand gate (Gate B) and the requirement that all outbound text is created as a draft (Gate D) rather than sent directly, ensuring a human-in-the-loop review.
[PERSISTENCE]: The skill uses CronCreate to establish a daily execution schedule. It also references the potential use of a launchd agent for persistent scheduling on macOS hosts (~/Library/LaunchAgents/com.tim.deal-momentum.plist). These mechanisms allow the skill to execute periodically without manual intervention.
[COMMAND_EXECUTION]: The skill utilizes execute_sql to perform complex queries against the HubSpot deals database and ask_agent to query unstructured activity history. While these are used for legitimate business analysis, they represent powerful tools for data access that rely on the agent's ability to maintain context and intent while constructing queries.

deal-momentum-analyzer-skill