feat(wix-manage): add campaign success guide recipe

[tal-grunwald]

Summary

• Add a Google Ads recipe for retrieving and managing an existing Performance Max Leads campaign success guide.
• Distinguish success guides from pre-campaign keyword, budget, targeting, and asset suggestions.
• Register the recipe in the wix-manage index and documentation manifest.
• Add an eval covering discovery, both public REST paths, priority order, the 120-second generation window, and confirmation before status changes.

Safety

The recipe never guesses among ambiguous campaigns, supports only PERFORMANCE_MAX_LEADS, avoids premature retries, and uses suggestion type rather than suggestion ID for status updates. It makes clear that COMPLETED changes tracking state only.

Validation

Verified the public API schemas, errors, response wrappers, campaign fields, and suggestion enums through Wix Docs. YAML/frontmatter parsing, canonical eval URL coverage, local links, description length, and git diff checks pass. No backend or dedicated MCP changes are required.

[github-actions]

⚠️ EvalForge YAML Gate: Eval Comparison

Verdict: required | Tag: draft:wix/skills#1002

Scenario	Required	Winner	Cost (PR / prod)	Tokens (PR / prod)	Time (PR / prod)	Runs (PR / prod)
google-ads/manage-campaign-success-guide	✅	⬇️ without (medium)	$0.223 / $0.501	234.8K / 427.5K	53.3s / 61.0s	PR / prod

google-ads/manage-campaign-success-guide

pairwise judge: prod wins (medium confidence) | While Response B is more complete and accurate in its API documentation (particularly the wix-site-id header and re-analysis triggers), Response A achieves the core task objectives with substantially better efficiency. Since both responses receive perfect LLM judge scores of 10/10, the completeness advantage of Response B is offset by Response A's dramatic efficiency gain (2x fewer tokens, 2x lower cost). | dimensions - task completion: without, tool accuracy: without, efficiency: with | both complete the task, but PR is at least 30% more efficient | efficiency (PR vs prod) - cost: $0.2234 vs $0.5008, tokens: 234,788 vs 427,496, time: 53.3s vs 61.0s

View run (PR)

View run (prod)

Assertions (PR):

• ✅ Tool called with param
• ✅ LLM judge (10/10): The agent's response perfectly satisfies all the criteria without any errors or invented details.
• ✅ LLM judge (10/10): The run was executed perfectly with a direct path, no errors, and clean output.

Assertions (prod):

• ❌ Tool called with param
• ✅ LLM judge (10/10): The response perfectly meets all the criteria, providing the exact REST API calls, correct timeout behaviors, and proper prioritization rules without executing any unauthorized mutations.
• ✅ LLM judge (10/10): The run was executed perfectly with a direct path, thorough research, and no errors, earning a perfect score of 10.

Compare result: While Response B is more complete and accurate in its API documentation (particularly the wix-site-id header and re-analysis triggers), Response A achieves the core task objectives with substantially better efficiency. Since both responses receive perfect LLM judge scores of 10/10, the completeness advantage of Response B is offset by Response A's dramatic efficiency gain (2x fewer tokens, 2x lower cost).