clawdbot/docs/refactor/web-provider-split.md

Web Provider Refactor (Nov 26, 2025)

Context: src/provider-web.ts was a 900+ line ball of mud mixing session management, outbound sends, inbound handling, auto-replies, and media helpers. We split it into focused modules under src/web/ and adjusted tests/CLI behavior.

What changed

• New modules: session.ts, login.ts, outbound.ts, inbound.ts, auto-reply.ts, media.ts; barrel remains src/provider-web.ts.
• CLI adds warelay logout to clear ~/.warelay/credentials; tested in src/web/logout.test.ts.
• Relay now exits instead of falling back to Twilio when the web provider fails (even in --provider auto), so outages are visible.
• Tests split accordingly; all suites green.
• Structured logging + heartbeats: web relay now emits structured logs with runId/connectionId plus periodic heartbeats (default every 60s) that include auth age and message counts.
• Bounded reconnects: web relay uses capped exponential backoff (default 2s→30s, max 12 attempts). CLI knobs --web-retries, --web-retry-initial, --web-retry-max, --web-heartbeat and config web.reconnect/web.heartbeatSeconds tune the behavior.
• Backoff reset after healthy uptime; logged-out state still exits immediately.

How to use

• Link: warelay login --provider web
• Logout: warelay logout (deletes ~/.warelay/credentials)
• Run relay web-only: warelay relay --provider web --verbose

Follow-ups worth doing

• Document the new module boundaries in README/docs; add a one-liner explaining the no-fallback behavior.
• Add bounded backoff/jitter in monitorWebProvider reconnect loop with clearer exit codes. ✅
• Tighten config validation (mediaMaxMb, etc.) on load. ✅ (schema now includes web.* knobs)
• Emit structured logs for reconnect/close reasons to help ops triage (status, isLoggedOut). ✅
• Add quick troubleshooting snippets (how to read logs, restart relay, rotate creds).