Report Builder

Create, configure, and manage automated reports via the Studio Chat Reports API. Reports are executed by SAMI in a sandbox with the data-expert skill — this skill focuses on defining reports, not executing them.

Key Terminology

Assistants and playbooks are the same concept. Users say "assistant" — the API uses "playbook." The playbook_base_ids field in report configuration filters by assistant.

Setup

Uses the same environment variables as other Studio Chat skills:

export STUDIO_API_TOKEN="sbs_your_api_key_here"
export STUDIO_PROJECT_ID="your-project-uuid"

Scripts

reports.py — Reports API client

# List all reports
python3 scripts/reports.py list

# Get a specific report
python3 scripts/reports.py get <report_id>

# Create a report
python3 scripts/reports.py create --name "Weekly CX Report" \
  --instructions "Analyze conversations..." \
  --playbooks PLAYBOOK_BASE_ID_1,PLAYBOOK_BASE_ID_2 \
  --schedule manual --window 7 \
  --slack "#reports" --email "user@company.com,boss@company.com"

# Update a report
python3 scripts/reports.py update <report_id> --name "New Name" --slack "#new-channel" --email "new@company.com"

# Delete a report
python3 scripts/reports.py delete <report_id>

# Trigger a manual run
python3 scripts/reports.py run <report_id> [--window 7]

# List runs for a report
python3 scripts/reports.py runs <report_id>

# Get run status and logs
python3 scripts/reports.py run-status <run_id>

Important: Do Not Run Unless Asked

Creating a report does NOT mean running it. By default, only create the report definition. Do NOT trigger a run unless the user explicitly asks to execute or run it.

• "Create a weekly report" → create the definition only
• "Create and run a report" → create + trigger run
• "Generate a report right now" → create + trigger run
• "Set up a report for me" → create the definition only

Workflows

Create a Report Definition

Default workflow — create the definition without executing:

1. Discover — List playbooks to identify which assistants to include
2. Design — Craft instructions based on what the user wants to analyze
3. Create — Create the report definition
4. Confirm — Tell the user the report was created and how to run it (from the UI or by asking)

One-Off Report (only when user asks to run)

Only when the user explicitly asks to generate/run/execute a report now:

1. Create — Create a manual report definition
2. Run — Trigger a manual run
3. Monitor — Poll run-status until completed (check every 15–20 seconds)
4. Deliver — Share the result: the report is viewable in the Reports UI

# Create + run (only when explicitly requested)
python3 scripts/reports.py create --name "Ad-hoc: Last 7 days" \
  --instructions "..." --window 7 --playbooks BASE_ID
# Note the report ID from output, then:
python3 scripts/reports.py run REPORT_ID --window 7
# Monitor:
python3 scripts/reports.py run-status RUN_ID

Recurring Report

When the user wants a scheduled report:

1. Discover — List playbooks
2. Design — Craft instructions based on what the user wants
3. Create — Create with --schedule cron --cron "EXPRESSION" and optionally --slack "#channel"
4. Confirm — Tell the user the report is scheduled. Do NOT trigger a test run unless asked.

Report Configuration

Fields

Field	Required	Description
name	Yes	Display name for the report
instructions	Yes	What SAMI should analyze and report on
schedule_type	No	manual (default) or cron
cron_expression	No	Cron schedule (min interval: 30 min). E.g. 0 9 * * 1-5
playbook_base_ids	No	Filter to specific assistants. Empty = all
time_window_days	No	Lookback window for manual reports (default: 7)
slack_channel	No	Slack channel for PDF delivery. E.g. #reports
email_recipients	No	List of email addresses for PDF delivery. E.g. ["a@b.com"]

Schedule Presets

Description	Cron
Weekdays at 9 AM (ART)	0 12 * * 1-5
Weekly Friday 3 PM (ART)	0 18 * * 5
Weekly Monday 9 AM (ART)	0 12 * * 1
Daily at 6 PM (ART)	0 21 * * *
Monthly on the 1st (ART)	0 12 1 * *

Note: Cron reports auto-calculate the time window from the interval (daily = 1 day, weekly = 7 days). Manual reports use the time_window_days field.

Writing Good Instructions

The report sandbox already has the data-expert skill loaded. This means SAMI already knows how to fetch conversations, compute metrics, call the analytics API, and process data. Do not include API endpoints, scripts, or fetch instructions in the report instructions.

Instructions define what to analyze and what output to produce — never how to get the data.

Instruction Principles

1. Describe the analysis, not the implementation — "Show deflection rate by playbook" is correct. "Call GET /projects/.../analytics" is wrong — the data-expert skill handles that.
2. Be specific about metrics — Name the KPIs: deflection rate, sentiment distribution, handoff rate, response time, conversation volume
3. Specify comparisons — "Compare to the previous period" gives context to the numbers
4. Name the playbooks — Even if filtered via playbook_base_ids, mention them by name for clarity
5. Ask for actionable insights — "Identify top 3 issues" is better than "analyze everything"
6. Structure the output — "Start with key metrics, then breakdown by topic, then recommendations"
7. Never include technical details — No URLs, no script paths, no API tokens. The execution environment handles all of that.

Example Instructions

Daily Operations Report:

Analyze all conversations from the time window.

1. Start with key metrics: total conversations, deflection rate, average response time, sentiment breakdown.
2. Break down by playbook: show each assistant's volume, deflection rate, and sentiment.
3. Identify the top 5 topics customers asked about.
4. Flag any conversations with very negative sentiment — summarize what went wrong.
5. End with 3 actionable recommendations for the CX team.

Weekly Performance Summary:

Generate a weekly performance summary.

1. Key metrics overview with week-over-week comparison.
2. Deflection rate trend — is it improving?
3. Sentiment distribution: what percentage of conversations are positive/neutral/negative?
4. Top 10 topics by volume with sentiment breakdown per topic.
5. Handoff analysis: which topics cause the most escalations?
6. Knowledge base gaps: are there topics where the AI consistently fails?
7. Recommendations for playbook improvements.

Playbook-Specific Analysis:

Deep dive into the Cotizador assistant.

1. Volume and deflection rate for this playbook only.
2. Most common questions customers ask.
3. Conversations where the assistant gave incorrect pricing — flag for review.
4. Average conversation length and response quality.
5. Comparison to previous period: is performance improving?

API Reference

Full endpoint documentation: references/api-reference.md