venus

Kpi Summary

Query KPI metrics from Postgres. Supported `kpi_domain`: - `sales`: provisional KPI v0 metrics from sales tables. - `membership`: membership KPI rows from the configured public view. For `sales`: - Supported dimensions: `retailstore`, `manager`. - Supported sort_by: `payment_amount`, `net_sales`, `refund_rate`, `suspicious_transaction_count`. For `membership`: - Supported mode: `overview`, `anomaly_candidates`, `store_detail`. - Use `month` as `YYYYMM

Membership Anomaly Ranking

Rank stores by membership anomaly_score for one month. Use this tool for requests such as "이상탐지 수치가 높은 매장", "anomaly_score 상위 N개", or monthly membership anomaly rankings. Inputs: - `month`: required month in `YYYYMM` or `YYYY-MM` format. - `limit`: number of stores to return, from 1 to 100. Defaults to 10. - `min_sales_qty`: optional minimum monthly sales quantity. - `min_new_customer_count`: optional minimum new-customer count. The tool always uses the members

Store Lookup

Find store candidates by store code or Korean store name. Use this tool before KPI analysis when the user mentions a store by a partial or ambiguous name, such as `아라리오W`, `센터시티W`, or `한화갤러리아`. Returns candidate rows with: - `shop`: store code used by sales rows. - `shop_name`: store display name from the lookup table. - `manager_code`: department/manager organization code. Agent guidance: - If exactly one likely store is returned, use its `shop` value for

Store Kpi Summary

Rank store KPI summaries or analyze a specific store. Use this tool when the user asks about a named store's sales, refunds, net sales, refund rate, or anomaly-candidate pattern. This is better than `kpi_summary` for questions like "아라리오W가 이상하다" or "한화갤러리아센터시티W 매장 패턴 분석해줘". Parameters: - `start_date` and `end_date` are ISO dates; `end_date` is exclusive. - Prefer `shop` when `store_lookup` already identified the exact store code. - Use `retailstore_keyword` when

Store Kpi Trend

Return daily KPI trend rows for a specific store. Use this tool when the user says a particular store looks abnormal or asks for a pattern over time. It supports any store code or store-name keyword; it is not fixed to a specific store. Parameters: - `start_date` and `end_date` are ISO dates; `end_date` is exclusive. - Prefer `shop` after using `store_lookup` to identify the exact store code. - Use `retailstore_keyword` only when the user gave a name fragment and

Store Kpi Comparison

Compare a specific store against same-manager and overall store averages. Use this tool when the user says a store looks abnormal and asks whether the result is high, low, unusual, or different from a baseline. It supports any store code or store-name keyword; it is not fixed to a specific store. Returns selected store metrics plus: - same-manager store count and average net sales/refund rate/anomaly-candidate count. - overall store count and average net sales/refund ra

Store Anomaly Assessment

Assess whether a specific store shows abnormal KPI patterns. Use this tool when the user says a store is suspicious, abnormal, weird, or asks whether a named store has unusual sales/refund behavior. It supports any store code or store-name keyword; it is not fixed to a specific store. The tool returns: - `risk_level`: `low`, `medium`, or `high`. - `signals`: Korean evidence statements explaining why the risk level was assigned. - `recommended_follow_up`: next checks

Data Availability Summary

Summarize loaded sales-data availability in Postgres. Use this before analysis when the user asks what data is loaded, what date range is available, or whether the ETL/sample load looks usable. Returns min/max `sale_date`, an estimated row count, and recent loaded days with row/store counts. This checks data availability only; it does not read customer identifiers. Agent guidance: - Use this to choose valid test dates before calling KPI tools. - Explain that th

Store Lookup Coverage

Check whether sales store codes are matched to store display names. Use this when the user asks why store names are missing, why results show store codes, or whether the store lookup table covers the loaded sales rows. Returns matched/missing store counts, match rate, and example missing store codes for the selected period. Agent guidance: - `end_date` is exclusive. - If match rate is below 100%, tell the user some stores will fall back to `shop` code until t

Store Return Code Breakdown

Break down one store's transactions by return_cd. Use this when a user asks why a store looks abnormal, whether returns are concentrated in return codes, or how much of the activity is return_cd 77/80. Returns store code/name, return_cd, row count, amount, quantity, anomaly-candidate count, and row share percent. Agent guidance: - Prefer `shop` after `store_lookup` identifies the exact store. - Treat return_cd 77/80 as anomaly candidates only, not confirmed fraud.

Store Anomaly Day Detail

Drill into one store's suspicious or spike date. Use this after `store_kpi_trend` or `store_anomaly_assessment` identifies a high-refund or high-anomaly-candidate day. Returns aggregate counts, return-code breakdown, item breakdown, and a small transaction sample. Samples intentionally exclude raw customer identifiers. Agent guidance: - `sale_day` is an ISO date such as `2011-01-06`. - Prefer `shop` after `store_lookup`. - Use this for follow-up evidence, not a

Manager Anomaly Ranking

Rank stores within one manager/department organization by anomaly metrics. Use this when a user wants to compare a suspicious store against the same 담당조직 or asks which stores in a manager group have high refund rate, net sales, payment amount, or anomaly-candidate count. Supported sort_by: `payment_amount`, `net_sales`, `refund_rate`, `suspicious_transaction_count`. Agent guidance: - `manager_code` is the sales row department code by default. - Explain that ran

Store Item Anomaly Breakdown

Break down one store's anomaly candidates by item/product fields. Use this when the user asks which products, brands, colors, sizes, or barcodes are driving a store's abnormal refund/anomaly pattern. Returns brand, item, seq, color, size, barcode, row count, quantity, payment amount, refund amount, anomaly-candidate count, and refund rate. Agent guidance: - Prefer `shop` after `store_lookup`. - Treat return_cd 77/80 as anomaly candidates only. - Return Korean r

Store Customer Anomaly Summary

Summarize store anomaly candidates by anonymous customer segments. Use this when the user asks whether a store's anomaly pattern is concentrated by customer segment. This tool never returns raw customer identifiers. Returns sex/age segment aggregates, suspicious counts/amounts, suppressed distinct-customer counts for small groups, and repeat-customer candidate counts. Agent guidance: - Do not claim a customer is suspicious from this aggregate. - If `suppressed_

Capability Plan Executor

Resolve and execute an AgentforcePlanner schema 1.0 Capability Plan. The external Plan keeps stable business Capability names. This MCP-side PoC maps them to the current read-only analytics tools through an explicit allowlist, supports a no-call dry run, and returns a concrete Action Plan, sanitized execution trace, coverage limitations, and report-ready handoff. This tool does not create or upload a report, mutate Salesforce metadata, write to Slack, or perform databas

Slack Upload Image

Upload one or every Venus KPI preview PNG to Slack. Use this tool when the user asks to upload, post, attach, or share a chart or table image in Slack. Security and parameter rules: - Provide `image_url` for one image or `image_urls` for an ordered image set. - Every value must be an exact HTTPS image URL. Duplicate URLs are uploaded once. - For KPI ranking previews, pass the directly constructed `chart_image_url` or `table_image_url`. - For store trends, pass

Kpi Preview Link

Build KPI preview links for chart and table rendering. Use this tool when the user asks for a chart, graph, visualization, Slack preview, or shareable KPI result. The returned `preview_url` renders an HTML page with a KPI bar chart and detail table. `image_urls` contains every PNG in upload order. The returned `chart_image_url` renders a 1200x630 PNG chart, and `table_image_url` renders a separate PNG table. `image_url` is kept as a backward-compatible alias for `cha

Store Trend Preview Link

Build preview links for a specific store's daily KPI trend chart. Use this tool when the user asks for a chart, graph, visualization, Slack preview, or image for a specific store's abnormal pattern over time. Supported metrics: - `net_sales`: daily net sales. - `payment_amount`: daily payment amount. - `refund_amount`: daily refund amount. - `refund_rate`: daily refund rate. - `suspicious_transaction_count`: daily anomaly-candidate count. The returned `prev

Action Plan Executor

Validate and execute an AgentPlanner schema 1.0 MCP Action Plan. The full plan is validated against an explicit planner Action registry and each registered MCP tool's current input schema before any Action is called. Dependencies, output references, failure isolation, and sanitized step statuses are preserved. The legacy capability_plan_executor remains a separate public tool and cannot be selected as a child Action.