Signalpilot

SignalPilot — Governed AI Database Access

User-visible Progress Contract

SignalPilot work is often tool-heavy. Keep the user oriented while you use MCP tools:

• Before the first MCP call, write one short sentence explaining what you are checking and why.
• Before each major phase, write one short progress sentence for schema scouting, query validation, model/notebook edits, execution, error repair, and final verification.
• After a tool result changes the plan, finds a data-quality issue, or fails, summarize what changed and what you will do next.
• Do not expose private reasoning, chain-of-thought, raw JSON, or every tool parameter. Explain intent and findings in human terms.
• Do not leave the user watching only MCP calls. A good run has concise assistant text between groups of tool calls.
• Final answers should mention the evidence actually checked and any residual uncertainty.

MCP Tools

The SignalPilot MCP provides governed database access:

• query_database — governed read-only SQL execution
• validate_sql / explain_query / estimate_query_cost — pre-execution checks
• schema_overview / schema_ddl / schema_link — schema discovery
• describe_table / explore_table / explore_columns / explore_column — deep dives
• list_tables / get_relationships / find_join_path — structure and joins
• compare_join_types — JOIN impact analysis
• get_date_boundaries — date ranges across all tables
• check_model_schema / validate_model_output / audit_model_sources — dbt validation
• analyze_grain — cardinality and grain analysis
• debug_cte_query — step-through CTE debugging
• dbt_error_parser — parse dbt error output into structured info
• list_projects / get_project — dbt project management
• check_budget / connection_health / query_history — operational
• list_notion_integrations / notion_search / notion_fetch_page / notion_create_page — Notion integration

Local Scripts (via plugin)

For local dbt project work, use these standalone scripts:

• scan_project.py — scan a dbt project: models, stubs, deps, hazards, work order
• validate_project.py — run dbt parse and report structural errors

Use ToolSearch to discover additional tools as needed.

Available Skills

Load these skills as needed for specialized work:

dbt Projects

• /signalpilot-dbt:dbt-workflow — Load FIRST for any dbt project. Full 8-step workflow: scan, load skills, validate, discover macros, research, spec, write, verify.
• /signalpilot-dbt:dbt-write — Load at the writing step for SQL model rules
• /signalpilot-dbt:dbt-debugging — Load when dbt run/parse fails
• /signalpilot-dbt:dbt-date-spines — Load to fix current_date/now() hazards
• /signalpilot-dbt:dbt-testing — Load for unit tests / data tests
• /signalpilot-dbt:dbt-snapshots — Load for snapshots / SCD2
• /signalpilot-dbt:dbt-versioning — Load for model versioning
• /signalpilot-dbt:dbt-knowledgebase — Load to use the knowledge base in dbt work

SQL (load the one matching your database)

• /signalpilot-dbt:duckdb-sql — DuckDB-specific syntax and gotchas
• /signalpilot-dbt:snowflake-sql — Snowflake-specific patterns
• /signalpilot-dbt:bigquery-sql — BigQuery-specific patterns
• /signalpilot-dbt:sqlite-sql — SQLite-specific patterns

Domain knowledge (load the one matching the data)

• /signalpilot-dbt:domain-ecommerce, domain-financial, domain-healthcare, domain-hr, domain-marketing, domain-media, domain-product

Database branching

• /signalpilot-dbt:xata — Xata Postgres branches: connect, diff two branches, and check whether an upstream branch's schema change breaks dbt before it merges

Notion Context

• /signalpilot-dbt:notion-context — Gather business context from Notion. Load this skill when the user asks for it; it is separate from the dbt workflow.

General

• /signalpilot-dbt:sql-workflow — Structured query building and verification
• /signalpilot-dbt:knowledge-base — Propose and search reusable knowledge entries
• /signalpilot-dbt:write-report — Write a structured analysis report

Quick Start

For dbt projects: Load /signalpilot-dbt:dbt-workflow — it orchestrates the full lifecycle including scanning, mapping, writing, and verification.

For SQL queries: Load /signalpilot-dbt:sql-workflow + the SQL skill for your database engine.

For schema exploration: Use the MCP tools directly — schema_overview for a broad view, describe_table for column details, explore_table for sample data.