Add AWS Comprehend PII redactor for JS SDK

[GaruBrothers]

Implements the AWS Comprehend PII redaction utility requested in #290 for the JavaScript SDK.

/claim #290

What changed

• Added AwsComprehendRedactor under the AI package.
• Uses Amazon Comprehend DetectPiiEntities to detect PII spans.
• Redacts prompts using stable AWS offsets, including Unicode-safe code point offset conversion for JavaScript strings.
• Supports filtering by entity type and confidence score.
• Adds helpers for prompt strings and chat message arrays.
• Exports the redactor from @arakoodev/edgechains.js/ai.
• Adds mocked unit tests for offset redaction, filtering, chat message handling and emoji/code-point offsets.
• Adds a working example under examples/aws-comprehend-redaction.

Verification

npx jest src/ai/src/tests/awsComprehendRedactor.test.ts --runInBand
npx tsc -p tsconfig.json --noEmit
npx tsc -b

Results locally:

• awsComprehendRedactor.test.ts: 6 tests passed.
• tsc --noEmit: passed.
• tsc -b: passed.
• GitHub Actions js: passed.
• GitHub Actions CLAAssistant: passed.
• npm run build: not run successfully on Windows because the existing script uses Unix rm -rf.

Note

The real AWS call requires AWS credentials and region configured in the environment. Tests mock the Comprehend client so CI does not need AWS credentials.

[github-actions]

CLA Assistant Lite bot All contributors have signed the CLA ✍️ ✅

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: be6959bcce

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Convert Comprehend code-point offsets before slicing

When the prompt contains supplementary Unicode characters before the PII span, this slices at the wrong positions because Comprehend reports offsets in UTF-8 code points while JavaScript slice and text.length use UTF-16 code units. For example, an emoji earlier in the text shifts every following AWS offset by one code unit, so the redactor can leave part of an email/phone number unredacted or remove adjacent non-PII text. Convert code-point offsets to JS string indices before validating and slicing.

Useful? React with 👍 / 👎.

[GaruBrothers]

I have read the Arakoo CLA Document and I hereby sign the CLA

[GaruBrothers]

recheck

[GaruBrothers]

Fixed the Unicode offset issue in dd70025 by converting Comprehend code point offsets to JavaScript UTF-16 code unit indices before slicing. Added an emoji regression test. Local verification: awsComprehendRedactor.test.ts now passes 6/6 tests, and npx tsc -p tsconfig.json --noEmit passes.