easterbrooka/mike

A production-hardened fork deployed on AWS under a new product name, with substantive work on encryption, email parsing, auth, and security controls.

Active development on a live AWS deployment. The encryption work is multi-PR and methodical - additive primitives, then wire-format fixes, then call-site cutover - suggesting an operator taking incremental production steps rather than shipping a big-bang refactor. Email parsing work has real test coverage (16 cases for attachment inlining, 6 for the .msg parser fixes). The Entra SSO change is opinionated and hard to disentangle from the useful system-key-fallback piece. Several security changes (IDOR fix, project boundary enforcement, Phase 1 hardening) indicate active security attention. The rebrand to "Michelle" and the presence of an `amplify.yml` confirm this is a production product deployment, not an experimental fork.

View on GitHub →

easterbrooka's fork of willchen96/mike is a working deployment - the commits reflect an operator shipping a product on AWS ECS and Amplify under the name "Michelle," not someone exploring the codebase. The substantive work falls into two clusters.

The first is an envelope-encryption programme running across multiple PRs. Phase 1 hardened logging, transport, and object storage and decoupled download-link signing from the Supabase service-role key. Phase 2 landed the cryptography primitives - AES-256-GCM envelopes, a KMS wrapper, HMAC email lookup, per-tenant DEK table - as an additive schema change. A follow-up commit then fixed the wire-format bug that would have made every encrypted write fail (raw Node Buffers need hex-literal conversion before they reach PostgREST). The dual-read/dual-write call-site cutover is in a separate PR.

The second cluster is email-document handling. easterbrooka added .msg upload support, fixed two silent parser failures affecting Outlook-exported files (blank bodies, dropped forwarded-message attachments), and extended the LLM read path to inline attachment text from within .msg and .eml files - PDFs, DOCX, XLSX, and nested messages up to three levels deep.

Other notable changes: Microsoft Entra SSO replaces email/password login, with a system-level LLM key fallback that lets users without personal API keys use models configured at the deployment level. The project folder boundary was promoted to a security control. Several smaller fixes addressed drag-and-drop uploads, disabled-button tooltips, and an IDOR in tabular review document access.

What's in it

Rebrand to Michelle A new product name and the CORS plumbing of a fork preparing for a real launch.
Microsoft Entra SSO Sign-in moves to Microsoft's identity stack, with the firm's own AI keys as a fallback for users who don't bring their own.
Envelope-encrypted secrets API keys, model settings and shared-workflow records now live as encrypted blobs instead of plaintext in the database.
Email and spreadsheet intake Mike accepts plain text, `.eml`, Outlook `.msg` and Excel - and reads the attachments stapled to emails, not just the cover note.
Matter folder as a security boundary Project isolation is enforced rather than assumed, including a fix to a document-access hole in tabular review.
AWS deployment path An ECS Fargate route and S3 support let the fork run on Amazon's infrastructure as well as Cloudflare's.
Sharper everyday UX Drag-and-drop on project pages no longer kicks you out of the app, and disabled Create buttons finally explain themselves.

Direction

securitybrandinginfrastructure

Activity

Themed changes and pull requests touching this fork, newest first. Themed changes that haven't been turned into a public post yet still appear — they're real work even without a published writeup.

✅ #17 Claude/fix msg html body and inner msg +508 -47 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

✅ #16 Fix .msg body extraction + inner-msg attachment detection +159 -45 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

Two bugs surfaced once users uploaded real .msg files from Outlook:

✅ #15 Inline email-attachment text in read_document for .eml and .msg +588 -10 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

When the LLM reads a .msg or .eml, it now also gets the extracted text of each attachment (PDF, DOCX, TXT, XLSX, recursive EML/MSG up to 3 nesting levels deep) appended to the email body. Lets the LLM answer "summarise t…

✅ #13 fix(security): scope tabular-review document_ids by access (CWE-639) +97 -6 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

The tabular-review routes accept user-supplied document_ids in request bodies (POST /tabular-review, PATCH /:reviewId) and stale cell rows on byte-fetching paths (POST /:reviewId/regenerate-cell, POST /:reviewId/generate…

✅ #14 fix project folder boundary checks +28 -3 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

✅ #12 Claude/fix projects allowlist drift +10 -12 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

✅ #11 Accept .msg (Outlook) uploads alongside .eml +217 -10 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

Outlook saves emails as proprietary binary OLE compound documents (.msg), not RFC 822. Extends the .eml work to also handle .msg by adding a parser that maps @kenjiuno/msgreader output onto the same ParsedEml shape - so …

✅ #10 Fix workflow shares causing crash +7 -1 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

getAdminClient() was reading NEXT_PUBLIC_SUPABASE_URL - that's a Next.js frontend convention that the backend never had defined. The empty fallback made supabase-js throw "supabaseUrl is required" synchronously from insi…

✅ #9 Fix: dropping files on a project page no longer triggers Chrome's def... +95 -2 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

...ault file-open

✅ #8 Accept .txt, .eml, .xlsx uploads with LLM-readable extraction + previews +2,000 -28 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

## Summary

✅ #7 Encode Buffer values as \x-hex before they cross the supabase-js boun... +125 -20 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

...dary

✅ #6 Phase 2 cutover: dual-read/dual-write for envelope-encrypted secrets +264 -26 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

Wires the Phase 2 crypto modules into the request path. After this lands and the backfill runs, every read prefers ciphertext (with plaintext fallback for un-backfilled rows) and every write seals the value under the use…

✅ #5 Add Phase 2 envelope-encryption foundations (migration + crypto modules) +1,710 -414 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

Pure additions; no app behaviour change. The new code isn't imported by the running backend yet - the call-site dual-read/dual-write changes ship in a follow-up so they can be reviewed in isolation.

✅ #4 Phase 1: encryption hardening (logging, transport, secrets, storage) +2,796 -201 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

## Summary

✅ #3 docs: add safe local testing guide +89 -1 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

Cherry-picks upstream `adc2cf2` from `willchen96/mike` (PR #31, "docs: add safe local testing guide").

✅ #2 Add CLAUDE.md behavioral guidelines +65 2mo ago self by easterbrooka ↗ analysis ↗ GitHub

Adds the CLAUDE.md guide from forrestchang/andrej-karpathy-skills,

📝 easterbrooka cuts Mike loose from Cloudflare and onto AWS 3 commits 3mo ago infrastructuresecurity draft

A small, tightly-linked set of changes lets the fork run on Amazon's cloud without rewiring the backend.

📝 easterbrooka makes the grayed-out buttons explain themselves 3 commits 3mo ago workflow draft

When you can't click Create, the app now tells you exactly what's missing instead of leaving you to guess.

📝 Operational handoff doc removed from repo 1 commit 3mo ago minor change

📝 easterbrooka kills the crash on shared-project lists 1 commit 3mo ago multi-tenantworkflow draft

A two-line database fix that stops the app falling over whenever someone opens their list of shared projects.

Show 20 more

📝 easterbrooka renames Mike to Michelle and lets one backend serve many front ends 1 commit 3mo ago brandinginfrastructure draft

A cosmetic rebrand rides along with a quietly useful change to how the backend decides which sites it will talk to.

📝 easterbrooka swaps passwords for Microsoft sign-in 2 commits 3mo ago securityintegration draft

This fork drops email-and-password login in favour of Microsoft Entra, the identity system many firms already use to control who gets into what.

✅ #1 Add AWS S3 support and backend Docker image for ECS deployment +246 -20 3mo ago self by easterbrooka ↗ analysis ↗ GitHub

## Summary

📝 Fix projects allowlist drift 0 commits — minor change

📝 easterbrooka/mike: AWS S3 support and ECS-ready Dockerfile for the backend 0 commits — infrastructureintegration draft

easterbrooka reworked the storage layer so the backend can deploy against AWS S3 via an ECS task role, without touching R2 compatibility, then added a production-oriented Dockerfile for Fargate.

📝 Import Karpathy-style coding discipline guide 0 commits — minor change

📝 easterbrooka/mike: upstream safe-testing guide cherry-picked, service-role key removed from frontend env example 0 commits — securityinfrastructure draft

easterbrooka pulled upstream commit adc2cf2 from willchen96/mike to add a local testing guide and, more materially, remove the Supabase service-role key from the frontend environment example file.

📝 Phase 1 encryption hardening: logging, transport, secrets, and object storage 0 commits — securityinfrastructure draft

easterbrooka ran an encryption review and landed the changes that didn't require a key-management redesign first. The pass covers PII scrubbing from logs, transport hardening, download-link signing decoupled from the Sup…

📝 Envelope encryption primitives land; nothing uses them yet 0 commits — securityinfrastructure draft

easterbrooka added the full cryptography layer for envelope-encrypted API keys and workflow shares - KMS wrapper, AES-256-GCM envelope, HMAC email lookup - without wiring any of it into the running backend. The split is …

📝 easterbrooka/mike: Phase 2 crypto wired into live request path with dual-write rollout 0 commits — securityinfrastructure draft

easterbrooka flipped user API keys, model settings, and workflow-share records onto envelope-encrypted columns across every read and write path, with a dual-write strategy so rolling back to a previous image doesn't lose…

📝 Buffer-to-hex conversion fixes bytea writes through PostgREST 0 commits — securityinfrastructure draft

When easterbrooka's envelope encryption started writing ciphertext columns, the first production insert would have 500'd. `JSON.stringify` serialises a Node Buffer as `{"type":"Buffer","data":[...]}`, which PostgREST can…

📝 easterbrooka/mike: txt, eml, and xlsx uploads added with server-side parsing and inline viewers 0 commits — intakediscovery draft

easterbrooka extended the document intake beyond pdf/docx to cover plain text, raw email, and spreadsheets - wiring each new type through the server-side parser, the LLM read path, and a new inline viewer component.

📝 Drag-and-drop file upload now works across the entire project page 0 commits — workflowintake draft

Before this fix, dragging a file onto the project page in Chrome would navigate away or trigger a browser download. easterbrooka patched the drop handling so files land in the right folder regardless of where on the page…

📝 Fix admin Supabase client env var crashing /workflows 0 commits — minor change

📝 easterbrooka/mike: Outlook .msg uploads added alongside .eml 0 commits — intakediscovery draft

easterbrooka extended the existing .eml ingestion path to cover Outlook's binary .msg format, mapping both onto a shared parsed-email shape so the viewer, the read_document tool, and the structure-tree extraction see no …

📝 Project folder boundary promoted to enforced security control 0 commits — securitymulti-tenant draft

easterbrooka is treating the project folder as an enforced isolation boundary. For a legal-AI tool where each matter holds privileged documents, a UI grouping alone is not enough.

📝 Inline attachment text into the LLM payload when reading .msg or .eml 0 commits — chat-uiworkflow draft

easterbrooka extended the email-reading path so the model sees attachment content alongside the message body - no manual detach-and-reupload needed. Pure read-side enrichment; the UI and storage layer are unchanged.

📝 Two silent failures in Outlook .msg parsing: empty body and dropped embedded messages 0 commits — intakediscovery draft

Real Outlook-exported `.msg` files exposed two bugs in easterbrooka's extractor: the message body came back empty, and emails forwarded as attachments were silently dropped. Both are fixed here.

📝 Fix HTML body and inner-message rendering in the chat surface 0 commits — chat-ui draft

easterbrooka patched the message rendering path to fix how assistant turns compose HTML bodies and how nested sub-messages get displayed. No diff was captured, but the branch name points at both problems sharing a common…

📝 easterbrooka/mike: IDOR in tabular review document access closed 0 commits — securitydiscovery draft

easterbrooka patched a CWE-639 authorization bypass in the tabular review routes, where four endpoints accepted caller-supplied document IDs without verifying the caller had read access to the underlying documents.

Threads of work (detailed view)

23 threads have been distilled into posts.

easterbrooka cuts Mike loose from Cloudflare and onto AWS

A small, tightly-linked set of changes lets the fork run on Amazon's cloud without rewiring the backend.

infrastructuresecurity

3 commits · latest 3mo ago

easterbrooka/mike: storage layer generalized to support AWS S3, ECS-ready Dockerfile added

easterbrooka extended the backend storage client so it can talk to either Cloudflare R2 or native AWS S3 without code changes, then added a multi-stage Dockerfile aimed at ECS Fargate. Both pieces landed in PR #1.

infrastructure

3 commits · latest 3mo ago

easterbrooka makes the grayed-out buttons explain themselves

When you can't click Create, the app now tells you exactly what's missing instead of leaving you to guess.

workflow

3 commits · latest 3mo ago

Radix tooltip explains why the Create button is disabled

The "Review name" and "Workflow name" inputs in the creation modals are styled as large serif headings, so users miss them and assume the greyed-out Create button is broken. easterbrooka iterated through three commits to fix it, landing on a Radix tooltip as the final approach.

chat-uiworkflow

3 commits · latest 3mo ago

easterbrooka kills the crash on shared-project lists

A two-line database fix that stops the app falling over whenever someone opens their list of shared projects.

multi-tenantworkflow

1 commit · latest 3mo ago

easterbrooka renames Mike to Michelle and lets one backend serve many front ends

A cosmetic rebrand rides along with a quietly useful change to how the backend decides which sites it will talk to.

brandinginfrastructure

1 commit · latest 3mo ago

Fork rebranded to "Michelle"; CORS extended for multi-origin Amplify deploys

easterbrooka renamed the product from "Mike" to "Michelle" across user-facing strings and extended the backend CORS handler to accept a comma-separated list of origins. The rebrand is fork-specific; the CORS pattern is reusable.

brandinginfrastructure

1 commit · latest 3mo ago

easterbrooka swaps passwords for Microsoft sign-in

This fork drops email-and-password login in favour of Microsoft Entra, the identity system many firms already use to control who gets into what.

securityintegration

2 commits · latest 3mo ago

Microsoft Entra SSO replaces email login; system LLM keys make models available for all users

Two changes in one commit: email/password auth is out, Microsoft Entra OAuth is in, and a new `/system/llm-providers` endpoint lets users without personal API keys use whatever models the deployment has configured at the environment level.

securityintegration

2 commits · latest 3mo ago