* Add CI workflow and apply ruff formatting
* Disable cdk synth — email_processor uses pre-built package dir
The email_processor Lambda bundles deps into a gitignored package/
directory. cdk synth fails in CI without a build step to recreate it.
Disabling until packaging is standardized.
* Use CDK BundlingOptions for email_processor Lambda packaging
Replaces the pre-built gitignored package/ directory with CDK's
built-in bundling. Deps are now installed inside a Docker container
during cdk synth, so the build works identically locally and in CI.
Re-enables run-cdk-synth in the CI workflow.
Replaces extraction prompt with domain-specific rules: trade
classification taxonomy (23 categories), site_code skip list,
zip padding, revision email type, and structured extraction for
fiscal_year, trade, and coupa_category.
Handler changes:
- New "revision" email type overwrites existing PO via put_item
- enrich_parsed() adds top-level state, ship_to_raw, data_source
- pad_zip() zero-pads short zip codes (e.g., "7001" → "07001")
- Removed invoice_total/invoice_count (Payee Central only)
Web UI: added revision badge, new detail fields (site code, state,
trade, fiscal year, coupa category, data source), line item table
now shows Qty/Unit/Price columns, list view shows Site and Trade.
CDK: fixed StreamViewType to match deployed table (NEW_IMAGE).
README: documented PO record schema and revision flow.
* Add CI/CD pipeline and fix stack name to kebab-case
CodePipeline V2 (po-ingest-pipeline) triggers CodeBuild on push
to main, running cdk deploy via buildspec.yml. Stack name changed
from PoIngestStack to po-ingest to match naming conventions.
* Add RETAIN policy to Secrets Manager secret
Prevents the Anthropic API key from being deleted if the stack
is ever removed. Matches the RETAIN policy on all other stateful
resources (DynamoDB tables, S3 bucket).
When no site code is found via Claude extraction or regex cascade,
the Lambda now checks the PO address against a cold-start cache of
verified-sites (normalized street + zip match). If still no match,
the PO is written to a new pending-site-review table for manual
verification against Payee Central.
The Claude extraction prompt now explicitly asks for site_code (the
Amazon facility code) and structured ship_to address fields (street,
city, state, zip). The site-extractor Lambda prefers these direct
fields when available, falling back to regex for older PO records.
Enable DynamoDB Streams on purchase-orders table and add a site-extractor
Lambda that extracts Amazon facility codes and addresses from PO ship-to
data, upserting them into a new verified-sites table. Includes a backfill
script for existing POs and upgrades existing Lambdas to arm64 + 60-day
log retention.
Pin existing CloudFormation stack name via stack_name property so
the live stack is not affected. Construct ID now follows the org
kebab-case standard.
CDK stack with SES receipt rule, S3 bucket, email processor Lambda
(Claude-powered extraction), web UI Lambda with Function URL, and
DynamoDB for storage. Includes reprocessing script for missed emails.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
