ci: check nightly commit secrets and email metadata

.github/scripts/check-commit-emails.sh

@@ -0,0 +1,40 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+event=${1:?event required}
+base=${2:?base SHA required}
+head=${3:?head SHA required}
+[[ $event == pull_request || $event == push ]]
+[[ $base =~ ^[0-9a-f]{40}$ && $head =~ ^[0-9a-f]{40}$ ]]
+# Missing/deleted history must fail, never silently reduce the checked range.
+git cat-file -e "$base^{commit}"
+git cat-file -e "$head^{commit}"
+
+allowed() {
+  local noreply='^[A-Za-z0-9+_.-]+(\[bot\])?@users[.]noreply[.]github[.]com$'
+  [[ $1 == dimitrikennedy@gmail.com || $1 == noreply@github.com ||
+    $1 =~ $noreply ]]
+}
+
+options=()
+if [[ $event == push ]]; then
+  options+=(--first-parent)
+fi
+commits=$(git rev-list "${options[@]}" "$base..$head")
+failed=0
+for commit in $commits; do
+  author=$(git show -s --format=%ae "$commit")
+  committer=$(git show -s --format=%ce "$commit")
+  parents=$(git show -s --format=%P "$commit")
+  # Upstream merge commits can have upstream identities. Side-branch commits
+  # are excluded on pushes; ordinary first-parent commits are never exempt.
+  if [[ $event == push && $parents == *' '* ]] && ! allowed "$author"; then
+    continue
+  fi
+  if ! allowed "$author" || ! allowed "$committer"; then
+    # Do not reproduce a potentially private address in public CI logs.
+    echo "Commit $commit has an author or committer email outside the allowlist." >&2
+    failed=1
+  fi
+done
+exit "$failed"

.github/scripts/check-commit-emails.test.sh

@@ -0,0 +1,72 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+root=$(git rev-parse --show-toplevel)
+script="$root/.github/scripts/check-commit-emails.sh"
+mkdir -p "$root/.generated"
+fixture=$(mktemp -d "$root/.generated/commit-emails.XXXXXX")
+trap 'rm -rf "$fixture"' EXIT
+cd "$fixture"
+git init -q
+tree=$(git mktree </dev/null)
+good=dimitrikennedy@gmail.com
+bad=contributor@example.com
+
+commit() {
+  local author=$1 committer=$2
+  shift 2
+  GIT_AUTHOR_NAME=Example GIT_COMMITTER_NAME=Example \
+    GIT_AUTHOR_EMAIL="$author" GIT_COMMITTER_EMAIL="$committer" \
+    git commit-tree "$tree" "$@" -m fixture
+}
+
+expect() {
+  local expected=$1
+  shift
+  local actual=0
+  bash "$script" "$@" >result 2>&1 || actual=$?
+  if [[ $actual != "$expected" ]]; then
+    echo "Expected status $expected, got $actual: $*" >&2
+    cat result >&2
+    exit 1
+  fi
+  if grep -q "$bad" result; then
+    echo 'Disallowed email leaked into output' >&2
+    exit 1
+  fi
+}
+
+base=$(commit "$bad" "$bad")
+valid=$(commit "$good" "$good" -p "$base")
+expect 0 pull_request "$base" "$valid"
+expect 0 push "$base" "$valid"
+expect 0 pull_request "$base" "$base"
+for email in '123+example@users.noreply.github.com' 'example@users.noreply.github.com' '123+example[bot]@users.noreply.github.com' noreply@github.com; do
+  head=$(commit "$email" "$email" -p "$base")
+  expect 0 pull_request "$base" "$head"
+done
+for email in "$bad" 'example@users.noreply.github.com.invalid' '@users.noreply.github.com'; do
+  head=$(commit "$email" "$good" -p "$base")
+  expect 1 pull_request "$base" "$head"
+  expect 1 push "$base" "$head"
+  head=$(commit "$good" "$email" -p "$base")
+  expect 1 pull_request "$base" "$head"
+done
+invalid=$(commit "$good" "$bad" -p "$base")
+tip=$(commit "$good" "$good" -p "$invalid")
+expect 1 pull_request "$base" "$tip"
+expect 1 push "$base" "$tip"
+upstream=$(commit "$bad" "$bad" -p "$base")
+merge=$(commit "$bad" "$bad" -p "$valid" -p "$upstream")
+expect 0 push "$base" "$merge"
+expect 1 pull_request "$base" "$merge"
+merge=$(commit "$good" "$bad" -p "$valid" -p "$upstream")
+expect 1 push "$base" "$merge"
+merge=$(commit "$good" "$good" -p "$valid" -p "$upstream")
+expect 0 push "$base" "$merge"
+# PR base reachability excludes upstream history already merged into the base.
+topic=$(commit "$good" "$good" -p "$upstream")
+expect 0 pull_request "$merge" "$topic"
+expect 128 push 0000000000000000000000000000000000000000 "$valid"
+expect 1 pull_request invalid "$valid"
+echo 'Commit email range tests passed.'

.github/scripts/commit-hygiene.test.sh

@@ -0,0 +1,109 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+root=$(git rev-parse --show-toplevel)
+gitleaks=${1:?absolute path to verified gitleaks executable required}
+mkdir -p "$root/.generated"
+fixture=$(mktemp -d "$root/.generated/commit-hygiene.XXXXXX")
+trap 'rm -rf "$fixture"' EXIT
+export GIT_AUTHOR_NAME=Example GIT_COMMITTER_NAME=Example
+export GIT_AUTHOR_EMAIL=example@users.noreply.github.com
+export GIT_COMMITTER_EMAIL=$GIT_AUTHOR_EMAIL
+# An inert stand-in for the private denylist secret. It must never be printed.
+fixture_name=acme-fixture-client
+LEAK_DENYLIST=$(cat <<'EOF'
+[[rules]]
+id = "fixture-private-name"
+description = "Inert denylist fixture"
+regex = '''(?i)\bacme-fixture-client\b'''
+keywords = ["acme-fixture-client"]
+EOF
+)
+export LEAK_DENYLIST
+git init -q "$fixture/source"
+cd "$fixture/source"
+tree=$(git mktree </dev/null)
+base=$(git commit-tree "$tree" -m base)
+left=$(git commit-tree "$tree" -p "$base" -m left)
+right=$(git commit-tree "$tree" -p "$base" -m right)
+
+# An inert custom rule proves merge-patch coverage without storing a credential.
+cat >.gitleaks.toml <<'EOF'
+[[rules]]
+id = "merge-fixture"
+description = "Inert regression marker"
+regex = '''fixture-only-merge-marker'''
+EOF
+blob=$(printf 'fixture-only-merge-marker\n' | git hash-object -w --stdin)
+merge_tree=$(printf '100644 blob %s\tresolution.txt\n' "$blob" | git mktree)
+merge=$(git commit-tree "$merge_tree" -p "$left" -p "$right" -m merge)
+git update-ref refs/heads/fixture "$merge"
+# The old invocation misses the addition that exists only in the merge tree.
+"$gitleaks" git --redact --log-opts="$base..$merge" >"$fixture/old-scan.log" 2>&1
+status=0
+bash "$root/.github/scripts/scan-commit-secrets.sh" "$base" "$merge" "$gitleaks" >"$fixture/merge-scan.log" 2>&1 || status=$?
+[[ $status == 1 ]]
+grep -q 'leaks found:' "$fixture/merge-scan.log"
+bash "$root/.github/scripts/scan-commit-secrets.sh" "$base" "$left" "$gitleaks" >"$fixture/clean-scan.log" 2>&1
+
+# The private-name denylist fails closed without its secret.
+status=0
+LEAK_DENYLIST= bash "$root/.github/scripts/scan-commit-secrets.sh" "$base" "$left" "$gitleaks" >"$fixture/no-secret.log" 2>&1 || status=$?
+[[ $status == 1 ]]
+grep -q 'LEAK_DENYLIST repository secret is not set' "$fixture/no-secret.log"
+status=0
+LEAK_DENYLIST='not toml [[' bash "$root/.github/scripts/scan-commit-secrets.sh" "$base" "$left" "$gitleaks" >"$fixture/bad-secret.log" 2>&1 || status=$?
+[[ $status == 1 ]]
+grep -q 'Private-name denylist scan failed' "$fixture/bad-secret.log"
+
+# A denylisted name fails the check wherever it appears, and only the rule ID is printed.
+expect_denied() {
+  local label=$1 commit=$2 where=$3 status=0
+  bash "$root/.github/scripts/scan-commit-secrets.sh" "$base" "$commit" "$gitleaks" >"$fixture/$label.log" 2>&1 || status=$?
+  [[ $status == 1 ]] || { echo "denylist missed: $label" >&2; exit 1; }
+  grep -q "rule fixture-private-name matched .* in $where" "$fixture/$label.log"
+  if grep -qi "$fixture_name" "$fixture/$label.log"; then
+    echo "denylist output leaked the name: $label" >&2
+    exit 1
+  fi
+}
+name_blob=$(printf 'const owner = "%s";\n' "$fixture_name" | git hash-object -w --stdin)
+plain_blob=$(printf 'x\n' | git hash-object -w --stdin)
+expect_denied content "$(git commit-tree "$(printf '100644 blob %s\tsrc.ts\n' "$name_blob" | git mktree)" -p "$base" -m content)" patches
+expect_denied message "$(git commit-tree "$tree" -p "$base" -m "work for $fixture_name")" identifiers
+expect_denied path "$(git commit-tree "$(printf '100644 blob %s\t%s.md\n' "$plain_blob" "$fixture_name" | git mktree)" -p "$base" -m path)" identifiers
+expect_denied author "$(GIT_AUTHOR_NAME="$fixture_name" git commit-tree "$tree" -p "$base" -m author)" identifiers
+HEAD_REF="feature/$fixture_name" expect_denied ref "$left" identifiers
+bash "$root/.github/scripts/scan-commit-secrets.sh" "$base" "$left" "$gitleaks" >"$fixture/clean-denylist.log" 2>&1
+# Redaction commits that delete or rename away a denylisted path still pass.
+named=$(git commit-tree "$(printf '100644 blob %s\t%s.md\n' "$plain_blob" "$fixture_name" | git mktree)" -p "$base" -m named)
+redacted=$(git commit-tree "$(printf '100644 blob %s\tneutral.md\n' "$plain_blob" | git mktree)" -p "$named" -m redact)
+bash "$root/.github/scripts/scan-commit-secrets.sh" "$named" "$redacted" "$gitleaks" >"$fixture/redaction.log" 2>&1
+
+# Reproduce a nested gitlink without .gitmodules, as found in vendored history.
+git update-index --add --cacheinfo "160000,$base,.repos/example/submodules/dependency"
+git update-index --add --cacheinfo "100644,$blob,.github/fixture.txt"
+checkout_tree=$(git write-tree)
+checkout_commit=$(git commit-tree "$checkout_tree" -p "$base" -m checkout)
+git update-ref refs/heads/checkout "$checkout_commit"
+git checkout -q -f checkout
+status=0
+git submodule foreach --recursive ':' >"$fixture/full-checkout.log" 2>&1 || status=$?
+[[ $status != 0 ]]
+grep -q 'No url found for submodule path' "$fixture/full-checkout.log"
+
+git init -q "$fixture/fetched"
+cd "$fixture/fetched"
+git remote add origin "$fixture/source"
+git -c credential.helper= fetch -q --no-tags --no-recurse-submodules origin "$base" "$checkout_commit"
+git checkout -q --detach "$checkout_commit"
+[[ -f .github/fixture.txt ]]
+[[ $(git rev-parse --is-shallow-repository) == false ]]
+git cat-file -e "$base^{commit}"
+if git config --local --get-regexp 'credential|extraheader|sshCommand'; then
+  echo 'Checkout unexpectedly stored authentication configuration' >&2
+  exit 1
+fi
+bash "$root/.github/scripts/check-commit-emails.sh" pull_request "$base" "$checkout_commit"
+bash "$root/.github/scripts/scan-commit-secrets.sh" "$base" "$checkout_commit" "$gitleaks" >"$fixture/checkout-scan.log" 2>&1
+echo 'Merge secret scan, private-name denylist and anonymous checkout regressions passed.'

.github/scripts/scan-commit-secrets.sh

@@ -0,0 +1,64 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+base=${1:?base SHA required}
+head=${2:?head SHA required}
+gitleaks=${3:?gitleaks executable required}
+[[ $base =~ ^[0-9a-f]{40}$ && $head =~ ^[0-9a-f]{40}$ ]]
+git cat-file -e "$base^{commit}"
+git cat-file -e "$head^{commit}"
+
+status=0
+# Compare merges with each parent so conflict-resolution additions are scanned.
+"$gitleaks" git --redact --log-opts="-m $base..$head" || status=1
+
+# Private-name denylist. The rules live only in the LEAK_DENYLIST repository secret
+# (gitleaks TOML), so this public repository never contains the names it guards.
+# Missing or unusable rules fail closed. Only rule IDs and where they matched are
+# printed; gitleaks output is discarded because it can echo file paths or messages.
+denylist=${LEAK_DENYLIST-}
+if [[ -z ${denylist//[[:space:]]/} ]]; then
+  echo "::error title=Private-name denylist missing::The LEAK_DENYLIST repository secret is not set or is empty (forked PRs never receive it). Failing closed; a maintainer must add the secret." >&2
+  exit 1
+fi
+work=$(mktemp -d)
+trap 'rm -rf "$work"' EXIT
+(umask 077 && printf '%s\n' "$denylist" >"$work/denylist.toml")
+
+denylist_scan() {
+  local kind=$1
+  shift
+  local report="$work/$kind.json" code=0
+  "$gitleaks" "$@" --no-banner --redact --exit-code 0 --config "$work/denylist.toml" \
+    --report-format json --report-path "$report" >/dev/null 2>&1 || code=$?
+  if [[ $code != 0 || ! -s $report ]] || ! jq -e 'type == "array"' "$report" >/dev/null 2>&1; then
+    echo "::error title=Private-name denylist scan failed::gitleaks could not scan $kind with the LEAK_DENYLIST rules (exit $code); output suppressed. Check that the secret is valid gitleaks TOML." >&2
+    exit 1
+  fi
+  local hits
+  hits=$(jq -r --arg kind "$kind" \
+    'group_by(.RuleID)[] | "\(.[0].RuleID | gsub("[^A-Za-z0-9._-]"; "?"))\t\($kind)\t\(length)"' "$report")
+  if [[ -n $hits ]]; then
+    while IFS=$'\t' read -r rule where count; do
+      echo "::error title=Private-name denylist::rule $rule matched $count time(s) in $where" >&2
+    done <<<"$hits"
+    status=1
+  fi
+}
+
+# Added lines in every new commit, including merge resolutions.
+denylist_scan patches git --log-opts="-m $base..$head"
+
+# Identifiers that never appear in patches: commit messages, author and committer
+# identities, paths a commit adds or renames to (removing a name is allowed) and the PR head ref.
+mkdir "$work/identifiers"
+while read -r sha; do
+  {
+    git show -s --format='%an%n%ae%n%cn%n%ce%n%B' "$sha"
+    git diff-tree -r -m --root --no-commit-id --name-only --diff-filter=d -M "$sha"
+  } >"$work/identifiers/$sha.txt"
+done < <(git rev-list "$base..$head")
+printf '%s\n' "${HEAD_REF-}" >"$work/identifiers/head-ref.txt"
+denylist_scan identifiers dir "$work/identifiers"
+
+exit "$status"

.github/workflows/commit-hygiene.yml

@@ -0,0 +1,64 @@
+name: Commit hygiene
+
+on:
+  pull_request:
+    branches: [hd/nightly]
+  push:
+    branches: [hd/nightly]
+
+permissions:
+  contents: read
+
+jobs:
+  scan:
+    runs-on: ubuntu-24.04
+    timeout-minutes: 10
+    env:
+      BASE_SHA: ${{ github.event.pull_request.base.sha || github.event.before }}
+      HEAD_SHA: ${{ github.event.pull_request.head.sha || github.event.after }}
+    steps:
+      # This public repository needs no checkout credentials. Avoid checkout's
+      # recursive auth cleanup, which fails on vendored gitlinks without URLs.
+      - name: Fetch full history without credentials
+        env:
+          GIT_CONFIG_NOSYSTEM: '1'
+          GIT_CONFIG_GLOBAL: /dev/null
+          GIT_TERMINAL_PROMPT: '0'
+        shell: bash
+        run: |
+          set -euo pipefail
+          [[ "$BASE_SHA" =~ ^[0-9a-f]{40}$ && "$HEAD_SHA" =~ ^[0-9a-f]{40}$ ]]
+          git init .
+          git remote add origin "${{ github.server_url }}/${{ github.repository }}.git"
+          git -c credential.helper= fetch --no-tags --no-recurse-submodules origin "$BASE_SHA" "$HEAD_SHA"
+          git checkout --detach "$HEAD_SHA"
+      - name: Validate range and commit emails
+        env:
+          EVENT_NAME: ${{ github.event_name }}
+        run: bash .github/scripts/check-commit-emails.sh "$EVENT_NAME" "$BASE_SHA" "$HEAD_SHA"
+      - name: Download and verify gitleaks
+        id: gitleaks
+        if: ${{ !cancelled() }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p "$RUNNER_TEMP/gitleaks"
+          cd "$RUNNER_TEMP/gitleaks"
+          curl --fail --silent --show-error --location --proto '=https' --tlsv1.2 \
+            https://github.com/gitleaks/gitleaks/releases/download/v8.30.1/gitleaks_8.30.1_linux_x64.tar.gz \
+            --output gitleaks.tar.gz
+          echo '551f6fc83ea457d62a0d98237cbad105af8d557003051f41f3e7ca7b3f2470eb  gitleaks.tar.gz' | sha256sum --check --strict
+          tar -xzf gitleaks.tar.gz gitleaks
+      - name: Test commit hygiene regressions
+        if: ${{ !cancelled() && steps.gitleaks.outcome == 'success' }}
+        run: |
+          bash .github/scripts/check-commit-emails.test.sh
+          bash .github/scripts/commit-hygiene.test.sh "$RUNNER_TEMP/gitleaks/gitleaks"
+      - name: Scan every new commit for secrets and private names
+        if: ${{ !cancelled() && steps.gitleaks.outcome == 'success' }}
+        shell: bash
+        env:
+          # gitleaks TOML rules; never committed. Unset or empty fails closed.
+          LEAK_DENYLIST: ${{ secrets.LEAK_DENYLIST }}
+          HEAD_REF: ${{ github.head_ref }}
+        run: bash .github/scripts/scan-commit-secrets.sh "$BASE_SHA" "$HEAD_SHA" "$RUNNER_TEMP/gitleaks/gitleaks"