fix(security): workflow permissions and two e2e regexes

.github/workflows/ci.yml

@@ -1,6 +1,6 @@
 name: CI
 
-# Kept deliberately cheap. Typecheck only, on the branches that matter — the
+# Kept deliberately cheap. Typecheck only, on the branches that matter: the
 # end-to-end suite needs a desktop session, a GPU and ffmpeg, so it runs before
 # a release on a real machine instead (see `npm run check:pre-release`).
 on:
@@ -16,6 +16,11 @@ concurrency:
   group: ci-${{ github.ref }}
   cancel-in-progress: true
 
+# Reads the code and nothing else. Without this the token gets the
+# repository's default, which can be write, for a job that only typechecks.
+permissions:
+  contents: read
+
 jobs:
   check:
     runs-on: windows-latest

.github/workflows/publisher-image.yml

@@ -7,7 +7,7 @@ name: Publisher image
 #
 # Only when the publisher itself changes. It used to run on every release tag,
 # which rebuilt an identical image for a release that had touched nothing in
-# `publisher/` — the app and the publisher are versioned and shipped
+# `publisher/`: the app and the publisher are versioned and shipped
 # separately, and there is no reason for one to drag the other along.
 on:
   push:
@@ -17,6 +17,12 @@ on:
       - '.github/workflows/publisher-image.yml'
   workflow_dispatch:
 
+# The image goes to Docker Hub with its own credentials, so the GitHub token
+# only ever reads the checkout. The layer cache does not need a scope here:
+# it authenticates with the runner's own token, not this one.
+permissions:
+  contents: read
+
 jobs:
   image:
     runs-on: ubuntu-latest
@@ -76,7 +82,7 @@ jobs:
             ${{ secrets.PUBLISHER_IMAGE }}:${{ steps.meta.outputs.version }}
             ${{ secrets.PUBLISHER_IMAGE }}:latest
           # Layers are cached between runs, so a change to one source file does
-          # not reinstall the dependencies again — which matters most for
+          # not reinstall the dependencies again, which matters most for
           # arm64, where every command runs under emulation. Scoped by name so
           # it cannot collide with another workflow's cache in this repository.
           cache-from: type=gha,scope=publisher

tests/e2e/compress.spec.ts

@@ -81,8 +81,9 @@ async function startFakePublisher(): Promise<FakePublisher> {
   // The real publisher refuses every write without this, so the fake one does
   // too: otherwise the test would pass whether or not the app sent it.
   const gate: express.RequestHandler = (req, res, next) => {
-    const given = /^Bearer\s+(.+)$/i.exec(req.header('authorization') ?? '')?.[1];
-    if (given !== TOKEN) {
+    // Compared whole rather than parsed: this is a test's fake publisher,
+    // and the exact header GoodBit sends is the thing being checked.
+    if (req.header('authorization') !== `Bearer ${TOKEN}`) {
       res.status(401).json({ message: 'Wrong or missing publish token.' });
       return;
     }

tests/e2e/publishGoodBits.spec.ts

@@ -47,8 +47,9 @@ async function startFakePublisher(): Promise<FakePublisher> {
   const TOKEN = 'a-test-publish-token';
 
   const gate: express.RequestHandler = (req, res, next) => {
-    const given = /^Bearer\s+(.+)$/i.exec(req.header('authorization') ?? '')?.[1];
-    if (given !== TOKEN) {
+    // Compared whole rather than parsed: this is a test's fake publisher,
+    // and the exact header GoodBit sends is the thing being checked.
+    if (req.header('authorization') !== `Bearer ${TOKEN}`) {
       res.status(401).json({ message: 'Wrong or missing publish token.' });
       return;
     }