proc-fixer

Proc Fixer (scripting flavor)

Repair a single converted mapping procedure — an Informatica mapping converted to a standalone Snowflake Scripting stored procedure — that still contains unsupported-transform stubs (!!!RESOLVE EWI!!!). Each materialized transformation is a block delimited by boundary markers; the fixer reconstructs each defective block's SQL from the source transformation node it was converted from, splices it back, then deploys, CALLs, and grades the procedure against the source-derived assertions until it creates clean and all assertions pass.

The fix is authored from source semantics, not guessed from the broken SQL, and the expected result the fix is graded against is computed by the test generator from the source — so the loop has an independent target.

Scope. This is the data-flow counterpart to orchestration-fixer (which repairs control-flow elements of a task graph) and the scripting-flavor analog of dbt-fixer (which repairs dbt models). In dbt mode a mapping becomes a model repaired by dbt-fixer; in scripting mode a mapping becomes its own stored procedure repaired here. It consumes the assertions authored by proc-test-gen — that pair owns the ARRANGE + ASSERT; this skill owns the ACT (the fix) and drives the repair loop.

Autonomous mode: When spawned as a team agent, skip all interactive stopping points, make reasonable defaults, and document assumptions in the completion report. When done, send_message to main with your results, and respond to any shutdown_request with type: "shutdown_response", approve: true.

Where this sits in the fix loop

block-locate  ── the defective block + its span/body (scripts/stabilization_tools.py block-locate, marker-scoped)
ewi-extract   ── the block's !!!RESOLVE EWI!!! / SSC-* defects (scripts/stabilization_tools.py ewi-extract)
proc-fixer    ── [THIS SKILL] reconstruct the fixed block SQL from the source transformation node
block-replace ── splice the fix back between the markers, integrity-checked (scripts/stabilization_tools.py block-replace)
deploy-and-call ─ deploy the one proc into the test schema + CALL it (scai code deploy-and-call)
grade         ── run the source-derived assertions against the populated target; PASS/FAIL feeds this loop