Add research digest: POST /research sends Gemini-powered email on any topic

[krMaynard]

Summary

• Adds POST /research endpoint that uses Gemini (the operator-configured GEMINI_MODEL) with Google Search grounding to research a configurable topic, then sends the summary directly to your Gmail inbox
• New assistant/researcher.py module with research_topic() — single Gemini call with live web search, 2048 token cap, rejects empty/blocked responses
• New send_message() in email_reader.py — uses existing gmail.modify scope (no re-auth needed), UTF-8 encoded
• RESEARCH_TOPIC env var controls what to research (e.g. "AI and LLM news this week"); missing topic returns 400
• Idempotent: the last sent (topic, date) is recorded in last_research.json in GCS, so a Cloud Scheduler retry after a timeout won't send the same digest twice on the same day (mirrors the /standup dedup pattern). Changing the topic sends a fresh digest
• README updated with Cloud Scheduler setup, env var docs, and architecture diagram

Test plan

• Set RESEARCH_TOPIC env var and hit POST /research locally — confirm email arrives in inbox
• Hit POST /research twice in a row — confirm the second returns {"skipped": true, "reason": "already_sent_today"}
• Deploy to Cloud Run and create a Cloud Scheduler job pointing at /research
• Check Cloud Run logs for Research email sent to ...

# Set topic
gcloud run services update assistant \
  --region us-central1 \
  --update-env-vars RESEARCH_TOPIC="AI and LLM news this week"

# Test locally
curl -X POST http://localhost:8080/research

# Add Cloud Scheduler job
gcloud scheduler jobs create http research-digest \
  --schedule="0 7 * * *" \
  --uri="$SERVICE_URL/research" \
  --http-method=POST \
  --oidc-service-account-email=cloud-scheduler-invoker@YOUR_PROJECT.iam.gserviceaccount.com \
  --location=us-central1 \
  --time-zone="America/Los_Angeles"

🤖 Generated with Claude Code

https://claude.ai/code/session_01CSnZNqUEh8uDyMCPDmzBTa

[gemini-code-assist]

Code Review

This pull request introduces a new research digest feature that uses Gemini with Google Search grounding to research a configured topic and email the summary directly to the user. Key changes include a new /research endpoint, a researcher module, and a helper to send emails immediately via the Gmail API. The review feedback highlights three important improvements: handling missing environment variables gracefully with a 400 Bad Request, adding defensive checks for empty or blocked Gemini responses, and explicitly specifying UTF-8 encoding when creating MIME text to prevent encoding errors.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[krMaynard]

Codex review

Fixed one concrete configuration regression in ba8d973: research digests now use GEMINI_MODEL (the same operator-controlled model as the rest of the service) instead of silently hardcoding gemini-2.0-flash. Added tests proving the configured model is passed through and empty responses remain rejected.

Validation: deterministic suite — 2 passed; git diff --check passed; no unresolved review threads.

One product-level blocker remains before merge: POST /research sends immediately but records no completed run. A Cloud Scheduler retry after a timeout can therefore send the same dated digest twice. The PR is also conflicting with main and should reconcile with the newer StateStore; that store is the natural place for a per-topic/per-date idempotency key. I did not guess those persistence semantics in this review.

[krMaynard]

Addressed the two remaining items from the Codex review:

• Merge conflict with main — rebased onto main (312e9d8). Resolved conflicts in main.py (both /availability-sync and /research now coexist) and README.md. PR is mergeable again.
• Double-send idempotency — POST /research now records the last sent (topic, date) in last_research.json in GCS and skips if today's digest for the current topic already went out, so a Cloud Scheduler retry after a timeout won't send twice. A changed RESEARCH_TOPIC still sends a fresh digest. This mirrors the existing /standup dedup pattern (there is no StateStore class on main — the pattern is the load_last_*/save_last_* GCS helpers, which is what I followed).

Full suite: 21 passed, 8 skipped (Gemini eval tests skip without a key locally).

---

Generated by Claude Code

[krMaynard]

Codex re-review at 2657ba0: the new read-then-send idempotency check could still duplicate a digest when overlapping Cloud Scheduler retries both read GCS before either request saved the record. I replaced it with an atomic per-topic/date GCS reservation using if_generation_match=0; only one request can create the marker. Definite pre-delivery failures release it for retry, while an ambiguous Gmail send failure retains it to favor duplicate prevention. Added focused atomicity/topic-scope/release tests and updated the documentation.

Validation: deterministic suite 23 passed; compileall and git diff --check passed.