The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content from call transcripts and CRM notes to generate customer-facing drafts and internal logs.
Ingestion points: External data is ingested via clari_get_call_summary (recording summaries, objections) and hubspot_get_deal in SKILL.md (Stage 1).
Boundary markers: The instructions lack explicit delimiters or 'ignore' instructions for the Clari-sourced summaries before they are used to populate drafts.
Capability inventory: The skill has the ability to create_draft in Gmail, create_event in Google Calendar, and write notes/tasks in HubSpot (hubspot_create_contact_note, hubspot_create_contact_task).
Sanitization: The skill mandates a 'Gate B' check using the check_my_copy tool on every customer-facing string before a draft is created, providing a layer of validation against malicious content.
[DATA_EXPOSURE]: The skill reads from and writes to specific hidden directories in the user's home folder for configuration and logging.
Evidence: The skill references paths such as ~/.claude/skills/epiphan-house-rules-skill/ for rules and writes analytics to ~/.claude/skill-analytics/ and ~/.claude/skill-runs/.
Context: While these appear to be standard paths for agent execution logging and cross-skill coordination within this specific environment, arbitrary file system writes represent a potential surface for persistence or data leakage if misconfigured.

post-demo-automation-skill