Deep-dive analysis of a single stored procedure. Resolves call graphs to base tables, classifies each statement as migrate or skip, produces a logic summary and migration guidance, and persists resolved statements to the procedure catalog file. This skill is called both directly by users and as a sub-agent by Skill Scoping Table.

/analyzing-object <schema.procedure>

Argument is the fully-qualified procedure name (e.g., dbo.usp_load_DimCustomer). The skill asks if missing. Only procedures are supported -- for views, functions, or tables, use Skill Listing Objects instead.

• manifest.json must exist in the project root. If missing, run Skill Setup DDL first.
• catalog/procedures/<name>.json must exist. If the object is a view, function, or table, the skill directs the user to /listing-objects show <name> instead.

uv run --project <shared-path> discover show --name <proc>

Returns refs, statements, classification, raw_ddl, params, and routing_reasons.

Checks the classification field from the catalog:

Read/write targets come from refs. Any ref that is a view, function, or procedure (not a base table) is resolved by running discover show on it to get its refs, following the chain until base tables are reached. The full lineage is presented as a tree:

silver.usp_load_DimCustomer  (direct writer)
  +-- reads: silver.vw_ProductCatalog (view)
  |     +-- reads: bronze.Customer
  |     +-- reads: bronze.Product
  +-- reads: bronze.Person
  +-- writes: silver.DimCustomer

Plain-language step-by-step description of what the procedure does, derived from raw_ddl.

Each statement is tagged as migrate or skip:

Deterministic procedures (classification: deterministic, no claude actions): all statements are already AST-classified. Persisted immediately with source: "ast". No additional user confirmation needed.

Claude-assisted procedures (classification: claude_assisted or statements containing action: "claude"):

1. Each claude statement is analyzed -- call graph followed, dynamic SQL resolved, classified as migrate or skip
2. Full resolved statement list presented for user confirmation
3. After confirmation, persisted with source: "llm". Each statement includes a rationale field.

No claude actions are written to catalog -- all must be resolved before persisting.

mkdir -p .staging
# Write statements JSON to .staging/statements.json
uv run --project <shared-path> discover write-statements \
  --name <procedure_name> --statements-file .staging/statements.json; rm -rf .staging

These fields are set by Skill Setup DDL and read by analyzing-object:

{
  "statements": [
    {
      "type": "Command",
      "action": "skip",
      "source": "ast",
      "sql": "TRUNCATE TABLE [silver].[DimCustomer]"
    },
    {
      "type": "Insert",
      "action": "migrate",
      "source": "ast",
      "sql": "INSERT INTO [silver].[DimCustomer] SELECT c.CustomerKey, c.FirstName, g.Region FROM [bronze].[Customer] c JOIN [bronze].[Geography] g ON c.GeographyKey = g.GeographyKey"
    }
  ]
}

{
  "statements": [
    {
      "type": "Command",
      "action": "skip",
      "source": "llm",
      "sql": "EXEC dbo.usp_truncate_staging @Table = 'DimCustomer'",
      "rationale": "Utility proc that truncates staging table. Operational overhead -- dbt handles table replacement via materialization."
    },
    {
      "type": "Insert",
      "action": "migrate",
      "source": "llm",
      "sql": "INSERT INTO [silver].[DimCustomer] SELECT ...",
      "rationale": "Core transformation that populates DimCustomer from bronze sources. This is the main SELECT that becomes the dbt model."
    }
  ]
}