E2E automated test suite for email agent

[FrederikHandberg]

Problem

The email agent (apps/email-agent/) test plan has 4 documented tests but none are automated. email-seen-ids.txt and the SEEN_FILE dedup logic is the most fragile part of the whole system and has had zero automated testing since it was written.

Currently: manual testing only. No regression safety net. Changes to the email agent could silently break deduplication, subject parsing, or capacity handling.

Fix

Create scripts/test-email-agent.sh that mocks the listen server and exercises:

1. New email → job submission: Mock a new message ID → verify POST to /jobs
2. Seen email → skipped: Same message ID again → verify no duplicate POST
3. Capacity full (429) → queue: Mock 429 response → verify email re-queued + capacity ack sent
4. Parse edge cases: Long subjects, HTML bodies, thread replies

Implementation:

• Use nc or a simple Python HTTP server to mock http://localhost:7600
• Run email agent against mock server
• Assert on mock server received requests

Acceptance Criteria

• 4 test cases automated in bash/python
• Tests run without a live email account (mock responses)
• Tests can be run via bash scripts/test-email-agent.sh
• CI: tests run on every PR that touches apps/email-agent/
• SEEN_FILE dedup tested with >100 IDs

Context

From Jensen limitation audit (2026-03-22). Workstream: WS-001.