
Manage Studio Chat project configuration — knowledge bases, playbooks, syncing, schedule, API tools, alerts, and trending topics. Use when asked to create, update, delete, or inspect KBs, playbooks, office hours, alerts, or any project settings. Also use to generate and browse trending topics analyses. Covers all CRUD operations via the Studio Chat API.
Create and configure automated reports in Studio Chat. Use when asked to set up a new report, schedule recurring reports, define report instructions, select which assistants/playbooks to include, configure Slack delivery, or manage existing report definitions. Expert at crafting report instructions that produce structured, high-quality output using the Block Kit format.
The proactive loop for shipping a behaviour change to an assistant — adding or changing a policy in the instructions or casuísticas (skills). Use when the request is "necesitamos que el asistente haga X", "agregá esta política", "cambiá el tono", "sumá una casuística para Y", or when a trend surfaced by data (trending topic, recurring handoffs, a monitor/alert) justifies an improvement. Drives the loop: clarify the policy → decide WHERE it lives (base instruction vs casuística vs KB vs example) → minimal draft → validate with in-memory overrides → ship via approvals → eval coverage. This is the feature/change-request counterpart to the quality-engineer skill (which is the reactive bug-report loop from a conversation ID).
Build and configure Studio Chat assistants — instructions, knowledge bases, skills, example blocks, API tools, toolkit actions (Slack), alerts, schedules, and trending topics. Use when asked to create, update, or manage any aspect of an assistant's configuration, including wiring up the template macros (pills) and the objects they reference. Covers all CRUD operations via the Studio Chat API.
Analyze customer conversation data, compute metrics, identify patterns, and generate reports using the Studio Chat Analytics API. Use when asked to analyze conversations, review performance, understand trends, examine deflection rates, sentiment distributions, handoff patterns, API tool usage, toolkit usage, resource analytics, sparklines, or any data analysis task involving platform activity.
Test and evaluate AI assistant behavior. Create test cases, run evaluations, analyze results, simulate conversations, and compare playbook versions. Also the end-to-end loop to debug and fix incorrect assistant behaviour starting from a conversation ID: root-cause it, validate a fix via overrides without saving a version, hand off to the human, and add regression evals. Use when asked to test an assistant, create QA scenarios, run evals, check assertion pass rates, verify assistant behavior, or investigate a conversation where the assistant misbehaved.