rule-engine-apply

Rule Apply

Apply matched rules to a single converted SQL file. For batch application (multiple files from propagation), see BATCH.md.

Inputs

• rules — matched rules array from search_rules (sorted by priority)
• file_path — path to the local SQL file to modify
• object_name — name of the object being migrated

Step 1: Sort by Priority

Process rules in priority order (ascending). Lower priority = runs first. This ensures simple regex fixes happen before structural AI rewrites.

Step 2: Apply Regex-Mode Rules

For each rule where replacement_mode = 'regex':

1. Read the current file content
2. Apply the regex substitution: find all matches of replacement_find and replace with replacement_replace
3. Write the updated content back to the file
4. Report what changed:

   Applied Replace ISNULL with COALESCE: 3 replacements made

If replacement_find does not match anything in the file (even though match_pattern did), skip silently — the pattern may have already been fixed.

Step 3: Apply AI-Mode Rules

For each rule where replacement_mode = 'ai':

1. Read the current file content
2. Read the rule's ai_context and examples
3. Use your judgment to identify and fix all instances of the pattern described
4. Present the diff to the user for approval before writing:

   EXEC to CALL — proposed changes:

   - EXEC schema.ProcName @p1, @p2
   + CALL schema.ProcName(:p1, :p2)
   

   Apply this fix?

5. If approved, write the updated file
6. If rejected, skip this rule

Step 4: Track

After each successful application, record it using the record_rule_application tool with rule_id set to the applied rule's ID and outcome set to "applied". Optionally include code_unit_name and file_path.

Output

Report back:

• Number of regex rules applied and total replacements made
• Number of AI rules applied (approved) and skipped (rejected)
• Whether the file was modified

Rules

• Always update the local file — do not deploy SQL directly without writing changes to the file first
• Regex rules do not need user approval — they are mechanical and deterministic
• AI rules always need user approval — present the diff before writing
• Re-read the file between rules — earlier fixes may affect later rule matches
• Track effectiveness — after testing confirms the fix works, call record_rule_application with outcome = "success"