Merge pull request #2506 from Molecule-AI/ci/secret-scan-required-and-precommit

secret-scan: align local pre-commit + extend drift lint (closes #1569 root)

.githooks/pre-commit

@@ -129,19 +129,57 @@ fi
 # ──────────────────────────────────────────────────────────
 # 6. Secrets: No tokens/keys in staged files
 # ──────────────────────────────────────────────────────────
+#
+# Pattern set MUST match .github/workflows/secret-scan.yml SECRET_PATTERNS
+# and molecule-ai-workspace-runtime/molecule_runtime/scripts/pre-commit-checks.sh —
+# .github/workflows/secret-pattern-drift.yml lints this invariant. Rebuilt
+# against canonical 2026-05-02 after #1569 Phase 1 discovery surfaced
+# real ghs_*/github_pat_* leaks that the prior pattern set
+# ('sk-ant-|sk-proj-|ghp_|gho_|AKIA|mol_pk_|cfut_') would have missed:
+# (a) it lacked ghs_ / ghu_ / ghr_ / github_pat_ / sk-svcacct- / sk-cp- /
+# xox[baprs]- / ASIA prefixes, (b) it skipped *.md and docs/* — but the
+# actual leaks lived in tick-reflections-temp.md, qa-audit-2026-04-21.md,
+# docs/incidents/INCIDENT_LOG.md.
+SECRET_PATTERNS=(
+  'ghp_[A-Za-z0-9]{36,}'           # GitHub PAT (classic)
+  'ghs_[A-Za-z0-9]{36,}'           # GitHub App installation token
+  'gho_[A-Za-z0-9]{36,}'           # GitHub OAuth user-to-server
+  'ghu_[A-Za-z0-9]{36,}'           # GitHub OAuth user
+  'ghr_[A-Za-z0-9]{36,}'           # GitHub OAuth refresh
+  'github_pat_[A-Za-z0-9_]{82,}'   # GitHub fine-grained PAT
+  'sk-ant-[A-Za-z0-9_-]{40,}'      # Anthropic API key
+  'sk-proj-[A-Za-z0-9_-]{40,}'     # OpenAI project key
+  'sk-svcacct-[A-Za-z0-9_-]{40,}'  # OpenAI service-account key
+  'sk-cp-[A-Za-z0-9_-]{60,}'       # MiniMax API key (F1088 vector — caught only after the fact)
+  'xox[baprs]-[A-Za-z0-9-]{20,}'   # Slack tokens (bot/app/user/refresh)
+  'AKIA[0-9A-Z]{16}'               # AWS access key ID
+  'ASIA[0-9A-Z]{16}'               # AWS STS temp access key ID
+)
 
 ALL_STAGED=$(git diff --cached --name-only --diff-filter=ACM || true)
 if [ -n "$ALL_STAGED" ]; then
   for f in $ALL_STAGED; do
-    # Skip binary, known safe files, hooks, docs, and markdown
-    if echo "$f" | grep -qE '\.png$|\.jpg$|\.ico$|\.woff|node_modules|\.lock$|\.githooks/|\.md$|docs/'; then
+    # Skip ONLY binary + lockfiles + the hook itself. Markdown +
+    # docs/* are NOT skipped — that was the bug (#1569 leaks were
+    # all in *.md). If a doc legitimately needs a token-shaped
+    # placeholder, use ghs_EXAMPLE_TOKEN_DO_NOT_USE — short enough
+    # to dodge the {36,} length suffix.
+    if echo "$f" | grep -qE '\.png$|\.jpg$|\.ico$|\.woff|node_modules|\.lock$|\.githooks/'; then
       continue
     fi
-    DIFF=$(git diff --cached "$f" 2>/dev/null | grep '^+' | grep -v '^+++' || true)
-    if echo "$DIFF" | grep -qE 'sk-ant-|sk-proj-|ghp_|gho_|AKIA[A-Z0-9]|mol_pk_|cfut_' 2>/dev/null; then
-      echo "❌ POSSIBLE SECRET in $f — do not commit API keys or tokens"
-      ERRORS=$((ERRORS + 1))
-    fi
+    DIFF=$(git diff --cached --no-color --unified=0 -- "$f" 2>/dev/null | grep -E '^\+[^+]' || true)
+    [ -z "$DIFF" ] && continue
+    for pattern in "${SECRET_PATTERNS[@]}"; do
+      if echo "$DIFF" | grep -qE "$pattern"; then
+        echo "❌ POSSIBLE SECRET in $f (matched: ${pattern})"
+        echo "   The actual matched value is NOT echoed here — round-tripping a"
+        echo "   leaked credential into scrollback widens the blast radius."
+        echo "   If false positive (test/docs example), use a short placeholder"
+        echo "   like ghs_EXAMPLE_TOKEN_DO_NOT_USE that doesn't satisfy the length."
+        ERRORS=$((ERRORS + 1))
+        break
+      fi
+    done
   done
 fi
 

.github/scripts/lint_secret_pattern_drift.py

@@ -41,6 +41,17 @@ CONSUMERS: list[tuple[str, str]] = [
     ),
 ]
 
+# In-repo consumers — paths read locally from the workflow checkout.
+# Read-from-disk avoids the staging→main lag that the URL fetcher
+# would hit (a freshly-edited canonical wouldn't yet be on the
+# consumer's default branch). Same drift semantics, no network.
+LOCAL_CONSUMERS: list[tuple[str, Path]] = [
+    (
+        ".githooks/pre-commit (molecule-core local hook)",
+        Path(".githooks/pre-commit"),
+    ),
+]
+
 # Matches the SECRET_PATTERNS=( ... ) array in either yaml-indented
 # (the canonical workflow's `run:` block) or shell-flat (runtime
 # hook) format. Patterns inside are single-quoted Bash strings; we
@@ -89,6 +100,27 @@ def main() -> int:
     print(f"canonical ({CANONICAL_FILE}): {len(canonical)} patterns")
 
     drift = False
+
+    # In-repo consumers first — these are read from the workflow's own
+    # checkout, so they never lag behind the canonical and a missing
+    # file IS a real error (not a fetch warning).
+    for label, path in LOCAL_CONSUMERS:
+        if not path.exists():
+            print(f"::error::{label}: file not found at {path}")
+            drift = True
+            continue
+        consumer = extract_patterns(path.read_text(), label)
+        missing, extra = diff_patterns(canonical, consumer)
+        if not missing and not extra:
+            print(f"  ✓ {label}: aligned ({len(consumer)} patterns)")
+            continue
+        drift = True
+        print(f"::error::DRIFT in {label}:")
+        for p in missing:
+            print(f"  -  missing from consumer: {p!r}")
+        for p in extra:
+            print(f"  -  extra in consumer (not in canonical): {p!r}")
+
     for label, url in CONSUMERS:
         try:
             content = fetch(url)

.github/workflows/secret-pattern-drift.yml

@@ -34,6 +34,7 @@ on:
       - ".github/workflows/secret-scan.yml"
       - ".github/workflows/secret-pattern-drift.yml"
       - ".github/scripts/lint_secret_pattern_drift.py"
+      - ".githooks/pre-commit"
   workflow_dispatch:
 
 # GITHUB_TOKEN scoped to read-only. The lint only does git checkout