GeezerVison

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Jul 7, 2026 2:07am

PR Summary

Medium Risk
Touches document ingestion (SSRF controls, file access) and accepts user-supplied AWS keys; Analyze ID adds PII handling, and the parse-route migration is a large behavioral swap though covered by tests.

Overview
Migrates Textract from hand-rolled SigV4 HTTP calls to @aws-sdk/client-textract, with shared helpers for document resolution (uploads, pinned URL fetch, S3 URI validation), async job polling, and SDK error mapping—including the multi-page PDF hint where async mode exists.

Adds two API surfaces and tools: Analyze Expense (sync AnalyzeExpense / async StartExpenseAnalysis + paginated GetExpenseAnalysis) and Analyze ID (sync AnalyzeID, optional back page), each returning camelCase-normalized payloads.

Extends textract_v2 with an Operation dropdown (defaults to existing analyze-document), routes to textract_parser_v2, textract_analyze_expense, or textract_analyze_id, operation-specific outputs, and back-of-ID inputs for ID verification. Zod contracts and the tool registry are updated; API route count baseline moves to 906.

^{Reviewed by Cursor Bugbot for commit 55cb4eb. Configure here.}

Greptile Summary

This PR replaces the hand-rolled AWS SigV4 signing implementation in the Textract parse route with @aws-sdk/client-textract, and adds two new operations — AnalyzeExpense (sync + async) and AnalyzeID (sync only) — behind a new operation dropdown in the textract_v2 block. The SDK migration is clean and removes ~240 lines of bespoke signing/polling code by extracting shared logic into a new shared.ts module.

• Shared utilities (resolveDocumentInput, pollTextractJob, mapTextractSdkError, parseS3Uri) are extracted into shared.ts, reducing duplication across the three route handlers, and all security guards (SSRF validation, path-traversal checks, S3 bucket name validation) are preserved in the new shared module.
• A function-typed condition (documentFieldCondition) ensures the front-document upload fields remain visible for analyze_id regardless of any stale processingMode value left over from a prior operation, addressing the previously reported UI bug.
• 29 unit tests are added covering route-level normalization functions, tool body builders, shared utilities (parseS3Uri, mapTextractSdkError, pollTextractJob), and response deserialization for all three operations.

Confidence Score: 5/5

Safe to merge — the SDK migration is a drop-in replacement for the hand-rolled SigV4 signer, all security guards are preserved in the new shared module, and the new operations follow the same patterns as the existing route.

The migration removes substantial bespoke code in favor of the official SDK and factors shared logic into a well-tested module. The documentFieldCondition function correctly handles the stale-processingMode scenario for analyze_id that was flagged in the previous review cycle. All three routes validate inputs at the contract layer and the route layer, SSRF and path-traversal guards are carried over intact, and 29 tests cover the key normalization and utility functions.

No files require special attention.

Important Files Changed

_{Reviews (6): Last reviewed commit: "chore(textract): drop redundant inline c..." | Re-trigger Greptile}

@greptile

@cursor review

@greptile

@cursor review

@greptile

@cursor review

@greptile

@cursor review

@greptile

@cursor review