Files
konstruct/.planning/phases/02-agent-features/02-03-SUMMARY.md
Adolfo Delorenzo 2dc94682ff docs(02-03): complete WhatsApp channel adapter plan
- Create 02-03-SUMMARY.md documenting WhatsApp adapter implementation
- Update STATE.md: advance progress to 56%, add 4 key decisions, record metrics
- Update ROADMAP.md: Phase 2 plan progress updated
- Mark CHAN-03, CHAN-04 requirements complete in REQUIREMENTS.md
2026-03-23 14:44:49 -06:00

160 lines
8.1 KiB
Markdown

---
phase: 02-agent-features
plan: "03"
subsystem: messaging
tags: [whatsapp, meta-cloud-api, hmac, webhook, minio, normalization, business-scoping, media]
requires:
- phase: 01-foundation
provides: "KonstructMessage model, channel gateway, normalize.py, resolve_tenant, check_rate_limit, is_duplicate/mark_processed"
provides:
- "WhatsApp Business Cloud API channel adapter (packages/gateway/gateway/channels/whatsapp.py)"
- "normalize_whatsapp_event() normalizing Meta Cloud API v20.0 payloads to KonstructMessage"
- "MediaAttachment and MediaType models in shared/models/message.py"
- "media: list[MediaAttachment] field added to MessageContent"
- "verify_whatsapp_signature() HMAC-SHA256 on raw body bytes"
- "verify_hub_challenge() for Meta webhook registration handshake"
- "is_clearly_off_topic() tier 1 keyword scoping gate"
- "build_off_topic_reply() canned redirect message builder"
- "send_whatsapp_message() and send_whatsapp_media() outbound delivery"
- "Media download from Meta API + MinIO storage with tenant-prefixed keys"
- "WhatsApp routes registered in gateway main.py"
- "whatsapp_app_secret, whatsapp_verify_token, MinIO settings added to shared/config.py"
affects:
- "02-agent-features/02-05 — outbound response routing in orchestrator tasks.py must route whatsapp channel through send_whatsapp_message"
- "02-agent-features — any plan touching gateway/normalize.py should import MediaAttachment/MediaType"
tech-stack:
added:
- "boto3 (not yet in pyproject.toml — added inline in media download function, needs uv add)"
- "httpx (already in gateway deps — used for Meta API calls)"
patterns:
- "Raw body read BEFORE JSON parsing for HMAC verification (prevent tampered-body attack)"
- "verify_whatsapp_signature returns raw bytes on success, raises HTTPException(403) on failure"
- "verify_hub_challenge returns challenge string on success, raises HTTPException(403) on failure"
- "normalize_whatsapp_event: meta-media://{media_id} placeholder URL before actual download"
- "Two-tier scoping: keyword gate (no LLM) -> LLM with system prompt scoping"
- "Always return 200 OK to Meta regardless of processing errors (Meta retry prevention)"
- "thread_id = sender wa_id for WhatsApp (no threading concept in WhatsApp)"
key-files:
created:
- packages/gateway/gateway/channels/whatsapp.py
- tests/unit/test_whatsapp_verify.py
- tests/unit/test_whatsapp_normalize.py
- tests/unit/test_whatsapp_scoping.py
modified:
- packages/shared/shared/models/message.py
- packages/shared/shared/config.py
- packages/gateway/gateway/normalize.py
- packages/gateway/gateway/main.py
key-decisions:
- "HMAC uses hmac.new() (not hmac.HMAC()) for timing-safe signature verification via hmac.compare_digest"
- "meta-media://{media_id} placeholder URL set at normalization time; actual download in adapter after tenant resolution"
- "thread_id set to sender wa_id — WhatsApp conversation scope is per-phone-number, not thread-based"
- "Always return HTTP 200 to Meta webhooks regardless of processing errors to prevent Meta retry storms"
- "Tier 1 scoping uses word-level tokenization with set intersection for keyword overlap check"
- "MinIO settings added to shared/config.py alongside WhatsApp credentials for use by all services"
- "boto3 required for MinIO access (S3-compatible client); needs uv add boto3 in gateway package before prod"
patterns-established:
- "WhatsApp adapter pattern: verify signature on raw bytes -> normalize -> tenant resolve -> rate limit -> dedup -> scope -> media -> dispatch"
- "Business-function scoping: tier 1 (keyword) rejects clearly off-topic without LLM; tier 2 (LLM) enforces via system prompt"
- "Canned redirect format: '{agent_name} is here to help with {topics}. How can I assist you with one of those?'"
requirements-completed:
- CHAN-03
- CHAN-04
duration: 7min
completed: 2026-03-23
---
# Phase 2 Plan 03: WhatsApp Channel Adapter Summary
**WhatsApp Business Cloud API adapter with HMAC-SHA256 verification, Meta v20.0 normalization, two-tier business-function scoping gate, and MinIO media storage**
## Performance
- **Duration:** ~7 minutes
- **Started:** 2026-03-23T20:36:19Z
- **Completed:** 2026-03-23T20:43:00Z
- **Tasks:** 2
- **Files modified:** 7 (4 created, 3 modified)
## Accomplishments
- WhatsApp Business Cloud API adapter with full inbound webhook processing pipeline
- HMAC-SHA256 signature verification on raw body bytes (timing-safe via hmac.compare_digest)
- normalize_whatsapp_event() for text, image, and document messages to KonstructMessage
- Two-tier Meta 2026 business-function scoping: keyword gate + LLM system prompt scoping
- MediaAttachment model with typed media fields added to shared message model
- Media download from Meta Graph API and storage in MinIO with tenant-prefixed keys
- Outbound delivery functions for text and media via Meta Cloud API
- 54 new passing tests (8 verify + 22 normalize + 14 scoping)
## Task Commits
Each task was committed atomically:
1. **Task 1: Media model extension, WhatsApp normalizer, and signature verification** - `370a860` (feat)
2. **Task 2: WhatsApp adapter with business-function scoping, media download/storage, and outbound delivery** - `6fea34d` (feat)
_Note: TDD tasks - tests written first then implementation_
## Files Created/Modified
- `packages/gateway/gateway/channels/whatsapp.py` - Full WhatsApp adapter: signature verification, hub challenge, webhook handler, scoping, media, outbound delivery
- `packages/gateway/gateway/normalize.py` - Added normalize_whatsapp_event() alongside existing Slack normalizer
- `packages/shared/shared/models/message.py` - Added MediaType enum, MediaAttachment model, media field on MessageContent
- `packages/shared/shared/config.py` - Added whatsapp_app_secret, whatsapp_verify_token, MinIO settings
- `packages/gateway/gateway/main.py` - Registered whatsapp_router, updated docstring
- `tests/unit/test_whatsapp_verify.py` - 8 tests for signature verification and hub challenge
- `tests/unit/test_whatsapp_normalize.py` - 22 tests for normalization (text, image, document)
- `tests/unit/test_whatsapp_scoping.py` - 14 tests for tier 1 scoping gate and canned reply
## Decisions Made
- HMAC uses `hmac.new()` with `hmac.compare_digest` for timing-safe comparison, preventing timing oracle attacks
- Normalizer sets `meta-media://{media_id}` as placeholder URL — actual download occurs in the adapter after tenant resolution, so the download has access to the access_token from channel_connections
- `thread_id = sender wa_id` — WhatsApp has no threading; conversation scope is per phone number
- Always return HTTP 200 to Meta even on processing errors — Meta retries all non-200 responses which can cause duplicate processing
- Tier 1 scoping uses word-level tokenization (text.split()) with set intersection, not substring matching, to avoid false positives
- MinIO settings co-located in shared/config.py so all services (gateway, orchestrator) can access MinIO
## Deviations from Plan
None - plan executed exactly as written.
## Issues Encountered
- Gateway package was not installed in the venv (only shared and orchestrator were). Running `uv sync --all-packages` resolved the import error. This is a Rule 3 (blocking) auto-fix.
## User Setup Required
The following environment variables must be added to `.env` before WhatsApp webhooks can receive real traffic:
```bash
WHATSAPP_APP_SECRET=<from Meta Developer Console>
WHATSAPP_VERIFY_TOKEN=<your chosen verification token>
MINIO_ENDPOINT=http://localhost:9000
MINIO_ACCESS_KEY=minioadmin
MINIO_SECRET_KEY=minioadmin
MINIO_MEDIA_BUCKET=konstruct-media
```
Also required before production: `uv add boto3` in `packages/gateway/` for MinIO uploads.
## Next Phase Readiness
- WhatsApp adapter is ready for tenant registration via channel_connections with phone_number_id as workspace_id
- Outbound response routing (LLM-generated replies via send_whatsapp_message) must be wired in Plan 02-05 (orchestrator tasks.py)
- MinIO bucket `konstruct-media` must be created before media messages can be processed
---
*Phase: 02-agent-features*
*Completed: 2026-03-23*