[Security 19] security posture: SECURITY.md, CodeQL, secret scan, Scorecard

.github/workflows/codeql.yml

@@ -0,0 +1,41 @@
+# CodeQL: static analysis for the whole tree (backend/ + frontend/).
+#
+# Both halves are interpreted TypeScript, so CodeQL's `build-mode: none`
+# extracts them straight from source — no npm ci, no build step, and no
+# autobuild guessing at the two-package layout. One analyze job covers
+# backend and frontend together. The weekly cron re-runs analysis on an
+# otherwise-quiet tree so newly shipped CodeQL queries still get applied.
+# Findings land in the Security tab (code scanning), not as CI noise.
+name: CodeQL
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+  schedule:
+    # Mondays 04:30 UTC.
+    - cron: "30 4 * * 1"
+
+permissions:
+  contents: read
+
+jobs:
+  analyze:
+    name: Analyze (javascript-typescript)
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      # Required to upload results to code scanning.
+      security-events: write
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+
+      - uses: github/codeql-action/init@a2983b8bed1923f44751c5c43237f479442827b3 # v3.37.4
+        with:
+          languages: javascript-typescript
+          build-mode: none
+
+      - uses: github/codeql-action/analyze@a2983b8bed1923f44751c5c43237f479442827b3 # v3.37.4
+        with:
+          category: "/language:javascript-typescript"

.github/workflows/gitleaks.yml

@@ -0,0 +1,45 @@
+# Secret scan: gitleaks over the full git history.
+#
+# Deliberately NOT gitleaks/gitleaks-action — that action requires a paid
+# GITLEAKS_LICENSE for organization-owned repos. The scanner itself is a
+# single MIT-licensed Go binary, so this downloads a pinned release and
+# verifies its sha256 (from the upstream release checksums file) before
+# running it. fetch-depth: 0 so every commit is scanned, not just the PR
+# head — a secret that was committed and later deleted is still leaked.
+# Known fake secrets (test fixtures, docs placeholders) are allowlisted
+# in .gitleaks.toml at the repo root; anything new fails the run.
+name: Secret scan
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+permissions:
+  contents: read
+
+jobs:
+  gitleaks:
+    name: gitleaks (full history)
+    runs-on: ubuntu-latest
+    env:
+      GITLEAKS_VERSION: 8.30.1
+      # sha256 of gitleaks_8.30.1_linux_x64.tar.gz, from
+      # gitleaks_8.30.1_checksums.txt on the upstream release.
+      GITLEAKS_SHA256: 551f6fc83ea457d62a0d98237cbad105af8d557003051f41f3e7ca7b3f2470eb
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          # Full history — gitleaks scans commits, not the working tree.
+          fetch-depth: 0
+
+      - name: Download and verify gitleaks
+        run: |
+          curl -sSfL -o gitleaks.tar.gz \
+            "https://github.com/gitleaks/gitleaks/releases/download/v${GITLEAKS_VERSION}/gitleaks_${GITLEAKS_VERSION}_linux_x64.tar.gz"
+          echo "${GITLEAKS_SHA256}  gitleaks.tar.gz" | sha256sum -c -
+          tar -xzf gitleaks.tar.gz gitleaks
+
+      - name: Scan full history
+        run: ./gitleaks detect --no-banner --redact

.github/workflows/scorecard.yml

@@ -0,0 +1,43 @@
+# OpenSSF Scorecard: automated checks of this repo's own security posture
+# (branch protection, token permissions, pinned dependencies, ...).
+#
+# Runs on pushes to main plus a weekly cron — Scorecard rates the repo, not
+# a PR's diff, so there is no pull_request trigger. publish_results: true
+# sends the score to the OpenSSF public API (which enables the README badge
+# and is what other tools query); that upload is signed via OIDC, hence
+# id-token: write. Results also land in the Security tab as SARIF.
+name: Scorecard
+
+on:
+  push:
+    branches: [main]
+  schedule:
+    # Mondays 05:30 UTC.
+    - cron: "30 5 * * 1"
+
+permissions: read-all
+
+jobs:
+  analysis:
+    name: Scorecard analysis
+    runs-on: ubuntu-latest
+    permissions:
+      # Needed to sign the published results (OIDC).
+      id-token: write
+      # Needed to upload SARIF to code scanning.
+      security-events: write
+      contents: read
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+        with:
+          persist-credentials: false
+
+      - uses: ossf/scorecard-action@4eaacf0543bb3f2c246792bd56e8cdeffafb205a # v2.4.3
+        with:
+          results_file: results.sarif
+          results_format: sarif
+          publish_results: true
+
+      - uses: github/codeql-action/upload-sarif@a2983b8bed1923f44751c5c43237f479442827b3 # v3.37.4
+        with:
+          sarif_file: results.sarif

.gitleaks.toml

@@ -0,0 +1,37 @@
+# gitleaks config for the Secret scan workflow (.github/workflows/gitleaks.yml).
+#
+# Default rules apply everywhere. The allowlist below suppresses EXACT known
+# fake secret values (anchored regexes tested against the found secret), not
+# files: every entry is one verified fixture string. A real credential added
+# anywhere — including to the files that hold these fixtures — would still
+# fail the scan, because its value cannot equal a known fake.
+#
+# Deliberately NOT path-based. Verified against gitleaks 8.30.1 by planting
+# a high-entropy secret in an allowlisted fixture file and scanning history:
+# a path-only allowlist swallows it (0 leaks reported), this config flags it.
+# Path entries also make `gitleaks dir` skip the whole file unread, so even
+# a path+value AND combination is a trap for anyone scanning the worktree.
+#
+# The scan covers full git history, so entries must keep matching fixtures
+# in old commits even if the files that hold them later move or change.
+
+[extend]
+useDefault = true
+
+[[allowlists]]
+description = "safeError unit-test fixtures: fake OpenAI-style key and fake token used to prove secrets get scrubbed from error output (backend/src/lib/__tests__/safeError.test.ts)"
+regexes = [
+    '''^sk-proj-abc123def456ghi789\.?$''',
+    '''^abcdef123456$''',
+]
+
+[[allowlists]]
+description = "Word add-in e2e fixture: mocked session token for the hermetic auth spec (word-addin/e2e/auth.spec.ts)"
+regexes = ['''^valid-jwt-123$''']
+
+[[allowlists]]
+description = "Supabase's public demo JWTs (iss: supabase-demo, HS256-signed with the published demo secret — the well-known local-dev anon/service_role keys from Supabase's own docs) in .env.example / docker-compose.yml history"
+regexes = [
+    '''^eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9\.eyJpc3MiOiJzdXBhYmFzZS1kZW1vIiwicm9sZSI6ImFub24iLCJleHAiOjE5ODM4MTI5OTZ9\.CRXP1A7WOeoJeXxjNni43kdQwgnWNReilDMblYTn_I0$''',
+    '''^eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9\.eyJpc3MiOiJzdXBhYmFzZS1kZW1vIiwicm9sZSI6InNlcnZpY2Vfcm9sZSIsImV4cCI6MTk4MzgxMjk5Nn0\.EGIM96RAZx35lJzdJsyH-qQwv8Hdp7fsn3W0YpN81IU$''',
+]

SECURITY.md

@@ -0,0 +1,45 @@
+# Security Policy
+
+## Supported versions
+
+| Version | Supported |
+| --- | --- |
+| tip of `main` | ✅ |
+| tagged releases (`v0.1.0`–`v0.4.0`) | ❌ |
+
+Security fixes land only on `main`; the existing tags are historical
+snapshots and do not receive backported fixes. If you are self-hosting,
+please update to the latest `main` before reporting — the issue may already
+be fixed.
+
+## Reporting a vulnerability
+
+Please report vulnerabilities privately through GitHub:
+
+**Security tab → Report a vulnerability** on this repository
+(GitHub's private vulnerability reporting).
+
+Please do not open a public issue for anything security-sensitive.
+
+This project has a solo maintainer. You can expect an acknowledgment within
+7 days; fixes are prioritized by severity after that.
+
+## Scope
+
+- Reports about the code, default configuration, and deployment guidance in
+  this repo are all in scope.
+- The **hosted service at `app.mikeoss.com`** (the cloud version offered from
+  the official Mike website) is in scope — report issues affecting it through
+  the same private channel above, and please keep testing non-destructive:
+  only accounts and data you own, no denial of service, no access to other
+  users' data beyond the minimum proof needed.
+- **Independent self-hosted installations** run by third parties are the one
+  exclusion: findings that only apply to how a specific outside operator has
+  deployed Mike (their infrastructure, their configuration) should go to
+  whoever operates that deployment.
+- Mike is an **LLM legal product**, so LLM-specific reports are explicitly
+  welcome: prompt injection (including via uploaded documents), getting the
+  model to ignore its guardrails, leaking another user's data or system
+  prompts through model output, and similar.
+- Secrets accidentally committed to this repository's history are also worth
+  a private report, even though CI runs a secret scanner.