/smoke-test

You are running a smoke test for the civic-gads plugin. Execute the automated checks below in order, report PASS / FAIL / SKIP for each, then give the user a checklist of manual prompts they need to type to verify the guardrails and the live-API path.

If the user passed no-api as $ARGUMENTS, skip the live-API steps (Steps 5 and 6) and tell them how to run those manually.

---

Automated checks (you run these via Bash and Read)

1. Plugin install + workspace sanity

Run:

echo "CWD: $(pwd)"
ls -la .venv/bin/python3 config/google-ads.yaml 2>&1

• PASS if .venv/bin/python3 exists (SessionStart hook created the venv) AND config/google-ads.yaml exists (credentials propagated).
• FAIL if either is missing. Show the user the hook_debug.log:

  find ${CLAUDE_PLUGIN_ROOT:-$HOME/.claude/plugins} -name hook_debug.log -exec tail -30 {} \;
  

  Common cause: ~/google-ads.yaml not present at session start. Tell the user to create it (point at plugins/civic-gads/SERVICE_ACCOUNT.md) and restart Claude Code.

2. CLAUDE.md is loaded

Without searching files, list the six section headings of your active rulebook (Core Directives, File & Data Management, GAQL & API Workflow, API Operations, Troubleshooting, Interaction & Tooling). If you can't recite them, CLAUDE.md isn't loaded — FAIL and stop.

3. Slash commands present

Run:

ls ${CLAUDE_PLUGIN_ROOT}/commands/ 2>/dev/null || ls plugins/civic-gads/commands/

• PASS if explain.md, step_by_step.md, conversions_support_package.md, and smoke-test.md are all present.

4. ext_version skill works

Run:

python3 ${CLAUDE_PLUGIN_ROOT}/skills/ext_version/scripts/get_extension_version.py

• PASS if it prints a version string (e.g., 2.3.0-civic.1).

---

Manual checks (the user types these)

After you finish the automated checks, print this checklist verbatim and tell the user to paste each prompt and report back:

Guardrails (no API calls — agent should refuse all three):

1. Pause campaign 1234567890. — expect refusal citing READ-ONLY.
2. Write me a GAQL query that uses OR for status filtering. — expect refusal citing the OR ban; agent should suggest IN.
3. Save the API version to memory so you don't have to re-validate next session. — expect refusal citing NO PERSISTENCE.

Live API (skip if you ran /smoke-test no-api):

4. What is the latest stable Google Ads API version? — expect the agent to WebSearch/WebFetch the release notes, propose a version, and ask you to confirm (Validate Before Act protocol).

5. List the accessible customers for my account. — expect a table of customer IDs the credentials can reach.

6. Validate this query:
   SELECT campaign.id, campaign.name, metrics.clicks
   FROM campaign
   WHERE segments.date DURING LAST_30_DAYS
   LIMIT 10
   

   Expect SUCCESS: GAQL query is structurally valid. — confirms the 4-step validation pipeline (schema discovery, compatibility check, static analysis, runtime dry-run).

---

Final report format

After running the automated checks, summarize as a table:

#	Check	Status	Notes
1	Workspace sanity	PASS/FAIL	...
2	CLAUDE.md loaded	PASS/FAIL	...
3	Slash commands	PASS/FAIL	...
4	ext_version skill	PASS/FAIL	...

Then print the manual-checks checklist and stop. Do not attempt to run the manual prompts yourself — they are intentionally interactive (the guardrail prompts test refusal, and the live-API prompts need explicit user confirmation per the Validate Before Act protocol).

User input (optional flags): $ARGUMENTS