The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest large amounts of untrusted data from multiple external sources to generate customer-facing content and internal briefings. This creates a significant surface for indirect prompt injection.
Ingestion points: Data is pulled from HubSpot contact/deal fields, Nooks call dispositions, Google Calendar event details, and Clari call summaries (Section 7).
Boundary markers: The instructions do not define any delimiters or specific 'ignore' instructions for the external content before it is processed by the model.
Capability inventory: The skill possesses the capability to generate and save Gmail drafts (create_draft) and send automated messages to Slack (slack_send_message).
Sanitization: There are no explicit instructions for sanitizing or escaping the data fetched from these external APIs before interpolation into prompts for email drafting or discovery question generation.
[DATA_EXPOSURE]: The skill instructions mandate the writing of an 'outcome sidecar' file to a path outside the repository workspace.
Evidence: The outcome_sidecar section specifies writing to ~/.claude/skill-analytics/last-outcome-morning-brief-skill.json. Writing data to hidden directories in the user's home folder is an unconventional operation that can lead to unintended data exposure or state tampering if the path is accessible to other processes.
[DATA_EXFILTRATION]: While listed as primary deliverable mechanisms, the skill automatically transmits aggregated CRM and productivity data to external services.
Evidence: The skill sends a summary of portfolio performance, dial activity, and deal alerts to Slack (slack_send_message) to the target ID U0AAJUZH2PK and creates drafts in a Gmail account (tkipper@epiphan.com). Although these are well-known services, the autonomous nature of the 'Tim is NOT present' run-mode means sensitive data is moved between platforms without real-time human oversight.

morning-brief-skill