ask

Altertable Ask

Central entry point for Altertable skills. Every Altertable task starts here. Match the user query to the best available skill and hand off.

Procedure

1. Read the user's query.
2. If using Altertable MCP tools, call initialize before any other Altertable tool so organization, environment, and knowledge context are correct.
3. Match the query against the routing table below, then apply the rules.
4. Activate the matched skill in your harness, passing the original query through so the matched skill has full context.
5. If the query is Altertable-related but no skill matches with confidence, use understanding-platform to orient the user.

For data questions, route to exploring-data when schema is unclear; otherwise route to querying-lakehouse.

Routing Table

Skill	When to route
exploring-data	Discover what data exists: catalogs, schemas, tables, columns, semantic models
querying-lakehouse	Answer questions that require querying lakehouse data using SQL
analyzing-funnels	Build or analyze a step-by-step conversion flow (drop-off between ordered events)
analyzing-web-traffic	Web analytics: pageviews, sessions, traffic sources, UTM, device, country breakdowns
analyzing-insights	Interpret an existing Insight or visualization the user is looking at
building-segments	Define or compare user cohorts by properties (not step-based)
forecasting-timeseries	Project future metric values or detect whether a change is within normal range
tracking-events	Work with tracked product analytics events, identities, or traits (querying or advising on instrumentation)
creating-insights	Create a new Insight that will be saved and visible to users
creating-discoveries	Create a discovery from a meaningful finding that should notify users and enter review
deciding-actions	Decide which insight or task type to use, or whether to create / update / skip a discovery
managing-discoveries	Review, approve, or reject existing discoveries and process user feedback on them
configuring-tasks	Set up a scheduled AI task (anomaly detection, forecast, monitor) that runs on a cron
using-memory	Search or create agent memories; create organization knowledge only when the user explicitly asks
evaluating-skills	Review or author agent skills themselves (skill structure, spec, quality)
understanding-platform	Explain Altertable concepts, architecture, or how agents work

When a skill is added, renamed, or removed from this repository, update this table in the same change.

Routing Rules

1. Single best match: pick one skill. Do not fan out.
2. Prefer the narrower skill: when two skills could match, prefer the more specific one.
3. Data-first when intent is vague: if the user wants to analyze data but does not specify how, start with exploring-data.
4. Fallback for unknown queries: if nothing matches with confidence, route to understanding-platform.
5. Pass context through: hand the original query to the matched skill.
6. Never invent a skill: only invoke skills that are actually installed.
7. Clarify before routing: if the query could reasonably mean different things, propose the most likely directions and let the user choose.