AI Travel Scout — test report

Date/time: 2026-06-25, ~20:43 CEST (build + end-to-end run)
Instance: n8n demo, http://[adres-twojej-instancji-n8n]:5678 (version 2.60.0)
Build mode: single agent (Opus), modular per-component testing, no Sonnet subagent fan-out (runtime had no Task tool; operator approved single-agent)
Workflows:
AI Travel Scout — Intake — id 2xESYQEBdmLOTLJE (5 nodes)
AI Travel Scout — Monitor Ranking Gmail — id IRcDHCb4oBeIpDKr (20 nodes)
Data Table: travel_scout_demo, id 0b0harbKvx3K3FeT, 32 columns
Test recipient (masked): to***twoj-testowy-adres@example.com
Export: outputs/travel-scout-final.workflow.json

Credentials used (no secrets)

OpenAI: OpenAI Demo Credential (gpt-4.1-mini) on all 3 AI Agent model nodes
SerpApi: SerpApi Demo Credential (HTTP Query Auth on the SerpApi node)
Gmail: Gmail Test Sender (OAuth2)

Test results

#	Test	Result	Evidence
1	Workflow structure	PASS	`n8n_validate_workflow` valid:true for both workflows
2	Credentials wired	PASS	OpenAI/SerpApi/Gmail bound on the right nodes
3	Intake (form → parse → insert)	PASS	execution `3034`, Data Table row id `2`, status `new`
4	SerpApi Google Hotels	PASS	execution `3038`, HTTP 200, 20 properties for "Barcelona hotels"
5	Ranking (TOP 3 + best)	PASS	best `Hotel 1898` (score 95), 3 alternatives, budget summary; table updated
6	Destination Guide	PASS	valid JSON: intro, facts, 4 activities, local tips
7	Travel Advisor + final body	PASS	full PL email body assembled (summary, hotel, pros/cons, TOP3, facts, activities, tips, plan)
8	Gmail send	PASS	execution `3038`, Gmail message id `19f00188f2f7a5eb`
9	Mail MCP verification	PASS	Gmail readonly MCP found message id `19f00188f2f7a5eb`, from/to `twoj-testowy-adres@example.com`, subject "Rekomendacja city break w Barcelonie - czerwiec 2026", body contains "Hotel 1898"
10	Final table update	PASS	row id `2`: `status=sent`, `recommendation_subject`, `recommendation_body`, `destination_guide`, `last_sent_at=2026-06-25T18:43:57Z`

Bugs found and fixed during the run

Intake form fields null — n8n_test_workflow form trigger maps fields by technical name field-0/1/2, not by label. First run errored at normalize ("Brak emaila"). Re-ran with field-N keys → PASS. In a real browser-submitted form the labels (email, travel_description, monitor_frequency) work directly.
Stale node references in Code nodes — initial Code nodes referenced course node names (Prepare query params, Extract hotels for AI, Parse AI ranking) but my nodes use prefixes (Set:/Code:). Patched via n8n_update_partial_workflow patchNodeField → PASS.
Gmail API disabled in Google Cloud — OAuth project [google-cloud-project-id] had Gmail API off (403 SERVICE_DISABLED). Enabled it in Cloud Console; retry → mail sent.

Executions

Intake error (field mapping): 3033
Intake success: 3034
Monitor error (node refs): 3035
Monitor error (node refs cascade): not retained
Monitor error (Gmail disabled): 3036, 3037
Monitor full success end-to-end: 3038

Risks / known issues

See travel-scout-known-issues.md.

Subagent reports

Built single-agent, but per-component evidence is split into the same module reports: - subagents/sonnet-intake-report.md - subagents/sonnet-monitor-ranking-report.md - subagents/sonnet-recommendation-gmail-report.md - subagents/sonnet-integration-qa-report.md - subagents/sonnet-mail-data-verification-report.md

Final verification

See opus-final-verification.md. Verdict: PASS. The mail arrived and was independently confirmed via Gmail MCP; Data Table is correctly updated; demo is showable.