AGENT SKILL

retail-discover-portfolio

Discover an unfamiliar retail data portfolio before per-table onboarding. Produces one metadata-only Layer-A survey across every reachable table, proposes non-critical domain and first-delivery scope decisions, delegates selected tables to retail-onboard-table for Layer-B profiling, and hands off to the existing business-knowledge-interview. Never samples source values, self-confirms a decision, or advances beyond the interview handoff.

License 许可
Apache-2.0
所属仓库
Kemetra/Seshat-BI
最近更新
2026-11-11
Stars
★ 2
Forks
0

01 前言 Front Matter

name
retail-discover-portfolio
description
Discover an unfamiliar retail data portfolio before per-table onboarding. Produces one metadata-only Layer-A survey across every reachable table, proposes non-critical domain and first-delivery scope decisions, delegates selected tables to retail-onboard-table for Layer-B profiling, and hands off to the existing business-knowledge-interview. Never samples source values, self-confirms a decision, or advances beyond the interview handoff.

03 SKILL.md 正文

retail-discover-portfolio

Use this skill when the owner has an unfamiliar database schema or file folder and needs a governed route from portfolio discovery to the business interview.

Boundary

This is an agent-conducted flow, not a new state engine or CLI workflow:

portfolio discovery -> domain -> scope -> selected-table onboarding -> interview handoff -> STOP

  • Layer A is the metadata-only portfolio survey in templates/portfolio-survey.md.
  • Layer B is the existing value-backed, per-table Source Ready profile owned by retail-onboard-table.
  • Never create a second profiler or author mappings/<table>/source-profile.md here.
  • Never select a scale-out route. Record scale evidence for the existing silver_gold_model_planning boundary to decide later.
  • Read the existing stage contracts and committed artifacts to derive exactly one next action. Create no run-state, projection, or routing file.

The golden reference shapes are:

  • tests/fixtures/portfolio-survey/db-schema/survey.md
  • tests/fixtures/portfolio-survey/file-folder/survey.md

1. Produce the Layer-A survey

For a database schema:

  1. Call seshat.portfolio_enumerate.enumerate_tables(schema). This helper is the only DB table-enumeration path. Do not issue a raw information_schema.tables query, and do not catch raw config, driver, or connection exceptions.
  2. If it returns an error, show only that redacted error. If no metadata is readable, STOP and name the unblock: configure the gitignored .env, install the matching retail DB extra, or grant metadata permission.
  3. For every returned table, read information_schema.columns plus declared PK/FK and catalog-estimate metadata through the same read-only boundary.

For a file folder, list every reachable CSV and Excel file and inspect only format/schema metadata. Do not read source values.

Fill one committed survey from templates/portfolio-survey.md:

  • include every reachable table; never choose a table-count or time cap;
  • include declared types, declared PK/FK metadata, catalog row estimates, and name/type-based date, PII-suspicion, grain, and structural-role hints;
  • label every inference candidate/hint, never a ruling;
  • for unavailable metadata, record [PENDING LIVE PROFILE] or needs_sample, the exact reason, and the enabling step;
  • never measure uniqueness, missingness, date spans, or returns population;
  • never include raw or masked samples, suspected-PII values, credentials, DSNs, or connection strings.

2. Propose a domain

Precondition: a committed, non-empty portfolio survey exists. The inherited domain_guess gate can pass with an absent store because its blocking_decision_categories is empty; therefore enforce this local precondition here. If missing, STOP: "portfolio survey missing; complete Layer-A discovery."

Load .seshat/semantic-decisions.yaml first and append a non-critical proposal:

  • decision_type: domain_classification (free-form, not a critical type)
  • status: proposed
  • confidence: low | medium | high (proposal confidence only)
  • proposed_by: agent
  • proposed_at: ISO-8601 timestamp
  • evidence: non-empty citations to survey facts

When evidence is ambiguous, record alternatives or "undetermined". Never approve, confirm, or add an approval-authority.yaml row. A named human may confirm through the existing low-risk batch path; follow spec 121's current status convention.

3. Propose a bounded first-delivery scope

Precondition: the domain proposal exists. If missing, STOP: "domain proposal missing; record a grounded domain guess."

Append a non-critical proposed scope decision with proposed_by, proposed_at, confidence, and non-empty survey + domain evidence. Record candidate tables, business questions, KPI names (not definitions), exclusions, and dependencies.

Bounding is deterministic:

  • honor an explicit owner-supplied scope limit;
  • otherwise prefer one coherent business process, one primary fact grain, and KPI names sharing one model boundary;
  • when metadata crosses processes/grains, present narrower coherent alternatives or record needs_user_input.

Describe the result categorically as coherent, cross-boundary, unresolved, or needs-user-input. Never store a numeric score, threshold, or rank. Partial acceptance creates a bounded new proposed record and marks the original superseded; never edit history in place.

For each selected table invoke retail-onboard-table. That existing skill alone produces Layer-B source-profile.md. Do not deep-profile inside the survey.

4. Hand off to the existing interview

Before handoff, require exactly the existing interview inputs:

  1. committed Layer-B per-table profiles for every selected table;
  2. the proposed scope;
  3. the existing Decision Store, loaded first and presented for confirmation or supersession without overwriting any record.

If a selected table lacks its Layer-B profile, STOP and name that table plus the retail-onboard-table unblock. Then invoke business-knowledge-interview; do not re-implement it and do not record an interview outcome here. Route KPI-meaning questions to the Retail KPI knowledge boundary.

At the interview handoff, STOP. This skill grants no approval and advances no readiness stage.

Resume and local-stop table

Committed state One next action
No survey Produce the Layer-A portfolio survey.
Survey, no domain proposal Record the grounded domain proposal.
Survey + domain, no scope Record the bounded scope proposal.
Scope, selected table lacks Layer-B profile Invoke retail-onboard-table for that table.
All Layer-B profiles + scope + loaded store Hand off to business-knowledge-interview, then STOP.
Request beyond interview handoff STOP; continue under the downstream stage contract, not this skill.

Every stop names the concrete missing artifact/decision and the action that unblocks it. Existing Decision Store records are never overwritten; reruns present them and changes use supersession.

04 同仓库更多 Skills

Kemetra
Kemetra/Seshat-BI

fake-skill

A fixture skill with declaring frontmatter, used as positive shipped evidence.

★ 2
Kemetra
Kemetra/Seshat-BI

fake-skill

A fixture skill with declaring frontmatter, used as positive shipped evidence.

★ 2
Kemetra
Kemetra/Seshat-BI

bare-skill-no-frontmatter

暂无简介

★ 2
Kemetra
Kemetra/Seshat-BI

fake-skill

A fixture skill with declaring frontmatter, used as positive shipped evidence.

★ 2
Kemetra
Kemetra/Seshat-BI

retail-kpi-knowledge

Retail KPI / metric-contract reasoning layer for BI agents in the Seshat BI project. Use when an agent must reason about the BUSINESS MEANING of a retail KPI before any implementation — defining a KPI in business terms, classifying additivity (fully/semi/non-additive), identifying grain, listing required source fields, resolving ambiguity (gross vs net, VAT, returns, cost method, same-store), choosing an MVP KPI pack, reviewing a metric contract for completeness, or preparing a clean handoff to an implementation layer. Also use it to classify KPI sufficiency or prepare an owner policy decision packet. This is a meaning/definition + review layer, not an executor: it does not write DAX/SQL/Python, grant readiness, approve policy, or design dashboards. See INDEX.md for live/planned routes; registry.yaml is the authoritative machine-readable inventory.

★ 2
Kemetra
Kemetra/Seshat-BI

bi-sql-knowledge

SQL reasoning, validation, reconciliation, transformation, diagnostics, and anti-pattern knowledge layer for Seshat BI agents. Use when a task involves reasoning about source SQL: source profiling, table grain, keys & uniqueness, joins & fan-out amplification, aggregation correctness, COUNT/NULL semantics, deduplication, validation queries, reconciliation queries, silver/gold transformation logic, window/date-time analytics, SQL anti-patterns, or basic SQL performance reasoning, or interpretation of a supplied sanitized PostgreSQL JSON execution plan. The SQL counterpart to bi-dax-knowledge. It is a thinking & validation layer, not a database executor, runtime validator, dbt/Dagster project, or automatic tuning service.

★ 2