The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources (email messages and calendar event descriptions) to update CRM records.
Ingestion points: Uses ~~почта (email) and ~~календарь (calendar) tools to gather context from external communications.
Boundary markers: The instructions do not define explicit delimiters for interpolated data, but they mandate the use of a specific "Cleanup Checklist" and mapping rules found in the reference files to guide the agent behavior.
Capability inventory: The agent has the ability to create contacts, link activities, and modify deal fields (amount, closing date, next steps) using the ~~crm tool.
Sanitization: The skill implements strict "Approval Gates," requiring the agent to present proposed changes to the user for explicit confirmation before any write operation is performed. This serves as the primary defense against the agent obeying malicious instructions embedded in external data.
[SAFE]: The skill contains no executable code or scripts, consisting entirely of instructional markdown and reference documentation.
[SAFE]: The instructions include a hard constraint against deleting any CRM records (contacts, deals, or activities), directing users to perform such actions manually within the CRM interface instead.
[SAFE]: Specific guidance is provided in reference/gotchas.md to prevent data corruption issues, such as performing case-insensitive email searches to avoid creating duplicate contact records.

crm-maintenance