secret-guard: no live credentials in bridged repos (hash-only findings)

New guard windy-git/secret-guard (warn-only) over EVERY text file: Telegram,
GitHub, AWS, Slack, Anthropic, OpenAI, Stripe live, Google API keys and
private-key blocks (scripts/secret_shapes.py, shared with the weekly public
scan). A finding carries "<kind> #<sha256[:8]>", never the value (house rule
10). Known fakes allowed BY HASH (ci/secret-guard-allow.yml). GUARDS_STATUS
gets a secrets column. Leak hunt 09-24: @Windy_0_bot token in a public fixture.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

api/tests/test_guards_report.py

@@ -57,7 +57,7 @@ def test_render_splits_lane_and_grant_counts():
     md = gr.render(res)
     assert "| ci-hygiene (house rule 6) | 1 | 1 | ❌ not yet |" in md
     assert "| compute-guard (Mind is the only door) | 0 | 0 | ✅ YES |" in md
-    assert "| windy-git | Windy Git | bbbbbbb | 0 | 0 | clean ✅ |" in md
+    assert "| windy-git | Windy Git | bbbbbbb | 0 | 0 | 0 | clean ✅ |" in md
     assert "| windy-pro | Windy Hub | aaaaaaa |" in md  # owner = session to message
     assert "(job reality-check)" in md
 

api/tests/test_secret_guard.py

@@ -0,0 +1,84 @@
+"""Secret guard: shapes, hash-only findings, allow by hash."""
+
+from __future__ import annotations
+
+import importlib.util
+import sys
+from pathlib import Path
+
+import pytest
+
+ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(ROOT / "scripts"))
+_spec = importlib.util.spec_from_file_location("secret_guard", ROOT / "scripts" / "secret_guard.py")
+sg = importlib.util.module_from_spec(_spec)
+sys.modules["secret_guard"] = sg
+_spec.loader.exec_module(sg)
+ss = sg.ss
+
+# Synthetic shapes only: none of these is a real credential.
+TG = "1234567890:" + "A" * 35
+CASES = [
+    ("telegram bot token", f"TELEGRAM_BOT_TOKEN={TG}"),
+    ("github token", "token = 'ghp_" + "a1" * 18 + "'"),
+    ("aws access key", "aws_access_key_id = AKIA" + "ABCDEFGHIJKLMNOP"),
+    ("slack token", "xoxb-" + "1234567890-abcdefghij"),
+    ("anthropic key", "ANTHROPIC_API_KEY=sk-ant-" + "x" * 30),
+    ("openai key", "OPENAI_API_KEY=sk-proj-" + "y" * 40),
+    ("stripe live key", "STRIPE=sk_live_" + "z" * 24),
+    ("google api key", "key=AIza" + "B" * 35),
+    ("private key block", "-----BEGIN OPENSSH PRIVATE KEY-----"),
+]
+
+
+@pytest.mark.parametrize("kind, text", CASES)
+def test_each_shape_is_found_and_only_its_hash_is_kept(kind, text):
+    hits = sg.scan_line("app.py", text)
+    assert [k for k, _ in hits] == [kind]
+    match = hits[0][1]
+    assert match.startswith(f"{kind} #") and len(match.rsplit("#", 1)[1]) == 8
+    # house rule 10: the value itself must never appear in a finding
+    secret = text.split("=", 1)[-1].strip(" '")
+    assert secret not in match
+
+
+@pytest.mark.parametrize("text", [
+    "sha512-" + "Q" * 86 + "==",           # lockfile integrity
+    "version: 12345678:abc",               # short, not a token
+    "sk-ant-short",                        # too short
+    "re_test_register_sends_verification", # windy-pro's fake Resend key
+    "ANTHROPIC_API_KEY=",                  # a name, not a value
+])
+def test_non_secrets_are_not_flagged(text):
+    assert sg.scan_line("x", text) == []
+
+
+def test_anthropic_key_is_not_double_counted_as_openai():
+    assert [k for k, _ in sg.scan_line("x", "sk-ant-" + "q" * 40)] == ["anthropic key"]
+
+
+def test_allow_is_by_hash_only():
+    F = sg.cg.Finding
+    fake = F("tests/t.py", 3, "telegram bot token", f"telegram bot token #{ss.h8(TG)}")
+    real = F("tests/t.py", 9, "telegram bot token", "telegram bot token #deadbeef")
+    kept = sg._drop_allowed("windy-chat", [fake, real], {"windy-chat": {ss.h8(TG)}})
+    assert kept == [real]
+    assert sg._drop_allowed("windy-mail", [fake], {"windy-chat": {ss.h8(TG)}}) == [fake]
+
+
+def test_allow_file_loads_and_needs_reasons(tmp_path):
+    assert sg.load_allow() == {} or isinstance(sg.load_allow(), dict)
+    bad = tmp_path / "a.yml"
+    bad.write_text("allow:\n  - repo: r\n    hashes: [abcd1234]\n")
+    with pytest.raises(ValueError):
+        sg.load_allow(bad)
+
+
+def test_block_and_warn(monkeypatch):
+    f = sg.cg.Finding("a.py", 1, "github token", "github token #abcd1234")
+    monkeypatch.setattr(sg, "MODE", "block")
+    assert sg.status_for([f], False)[0] == "failure"
+    assert sg.status_for([], False, grant=[f])[0] == "success"
+    monkeypatch.setattr(sg, "MODE", "warn")
+    state, desc, _ = sg.status_for([f], True)
+    assert state == "success" and desc.startswith("⚠ WARN (not blocking): 1 secret-shaped string in tree")

ci/secret-guard-allow.yml

@@ -0,0 +1,5 @@
+# Secret guard allow-list: KNOWN FAKE values that look like secrets (test
+# fixtures, docs). Allowed BY HASH (sha256[:8] of the value), never by path, so a
+# real secret in the same file still flags. Every entry MUST say why.
+# Owner: Windy Git lane (13); changes go through the orchestrator.
+allow: []

scripts/guards_report.py

@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-"""Live status of the repo guards (compute-guard + ci-hygiene) as one markdown page.
+"""Live status of the repo guards (compute-guard + ci-hygiene + secret-guard) as one markdown page.
 
 Scans every bridged repo's DEFAULT branch with both guards and renders what is
 left, per repo and owner lane. Findings in code Grant owns (ci/grant-owned.yml:
@@ -24,6 +24,7 @@ import yaml
 sys.path.insert(0, str(Path(__file__).resolve().parent))
 import ci_hygiene as hy  # noqa: E402
 import compute_guard as cg  # noqa: E402
+import secret_guard as sgd  # noqa: E402
 
 ROOT = Path(__file__).resolve().parents[1]
 OWNED = Path(os.environ.get("GRANT_OWNED", ROOT / "ci" / "grant-owned.yml"))
@@ -101,10 +102,11 @@ def scan(repo: str, owned: list[dict]):
         return None
     head = cg._git(bare, "symbolic-ref", "--short", "HEAD").strip()
     sha = cg._git(bare, "rev-parse", head).strip()
-    out = {"sha": sha, "compute": [], "hygiene": []}
+    out = {"sha": sha, "compute": [], "hygiene": [], "secrets": []}
     texts: dict[str, str] = {}
     for key, fs in (("compute", cg.check(repo, sha, head, True) or []),
-                    ("hygiene", hy.check(repo, sha, head, True) or [])):
+                    ("hygiene", hy.check(repo, sha, head, True) or []),
+                    ("secrets", sgd.check(repo, sha, head, True) or [])):
         for f in fs:
             job = None
             if "/workflows/" in f.path:
@@ -117,27 +119,29 @@ def scan(repo: str, owned: list[dict]):
 
 def render(results: dict) -> str:
     now = time.strftime("%Y-%m-%d %H:%MZ", time.gmtime())
-    lane = {k: 0 for k in ("compute", "hygiene")}
-    grant = {k: 0 for k in ("compute", "hygiene")}
+    lane = {k: 0 for k in ("compute", "hygiene", "secrets")}
+    grant = {k: 0 for k in ("compute", "hygiene", "secrets")}
     for r in results.values():
         for k in lane:
-            lane[k] += sum(1 for _, _, g in r[k] if not g)
-            grant[k] += sum(1 for _, _, g in r[k] if g)
+            lane[k] += sum(1 for _, _, g in r.get(k, []) if not g)
+            grant[k] += sum(1 for _, _, g in r.get(k, []) if g)
     L = [f"# Repo guards: live status (generated {now}; windy-git scripts/guards_report.py)",
          "_Default branches only. WARN-only today; the orchestrator says \"block\" per guard when its LANE column is 0. "
          "Grant-owned code (ci/grant-owned.yml) is listed separately and never holds up a block._", "",
          "| Guard | Lane-owned findings | Grant-owned (proposals) | Ready to block? |", "|---|---|---|---|",
          f"| compute-guard (Mind is the only door) | {lane['compute']} | {grant['compute']} | {'✅ YES' if lane['compute'] == 0 else '❌ not yet'} |",
          f"| ci-hygiene (house rule 6) | {lane['hygiene']} | {grant['hygiene']} | {'✅ YES' if lane['hygiene'] == 0 else '❌ not yet'} |",
-         "", "## By repo (lane-owned)", "| Repo | owner | head | compute | hygiene | first items |", "|---|---|---|---|---|---|"]
+         f"| secret-guard (no credentials in repos; hash only) | {lane['secrets']} | {grant['secrets']} | {'✅ YES' if lane['secrets'] == 0 else '❌ not yet'} |",
+         "", "## By repo (lane-owned)", "| Repo | owner | head | compute | hygiene | secrets | first items |", "|---|---|---|---|---|---|---|"]
     for repo, r in sorted(results.items()):
         c = [x for x in r["compute"] if not x[2]]
         h = [x for x in r["hygiene"] if not x[2]]
-        items = "; ".join(f"`{f.path}:{f.line}` {f.match}" for f, _, _ in (c + h)[:3]) or "clean ✅"
-        L.append(f"| {repo} | {OWNERS.get(repo, '?')} | {r['sha'][:7]} | {len(c)} | {len(h)} | {items} |")
+        s = [x for x in r.get("secrets", []) if not x[2]]
+        items = "; ".join(f"`{f.path}:{f.line}` {f.match}" for f, _, _ in (s + c + h)[:3]) or "clean ✅"
+        L.append(f"| {repo} | {OWNERS.get(repo, '?')} | {r['sha'][:7]} | {len(c)} | {len(h)} | {len(s)} | {items} |")
     L += ["", "## Grant-owned (windy-pro desktop app + its build jobs): proposals only, not blocking"]
-    g = [(repo, f, job) for repo, r in sorted(results.items()) for k in ("compute", "hygiene")
-         for f, job, own in r[k] if own]
+    g = [(repo, f, job) for repo, r in sorted(results.items()) for k in ("compute", "hygiene", "secrets")
+         for f, job, own in r.get(k, []) if own]
     L += [f"- {repo} `{f.path}:{f.line}`{f' (job {job})' if job else ''}: {f.match}" for repo, f, job in g] or ["- none"]
     return "\n".join(L) + "\n"
 

scripts/pr_status_bridge.py

@@ -372,6 +372,7 @@ def post_statuses(repo: str, sha: str) -> None:
 
 GUARD_CTX = "windy-git/compute-guard"
 HYGIENE_CTX = "windy-git/ci-hygiene"
+SECRET_CTX = "windy-git/secret-guard"
 
 
 def post_compute_guard(repo: str, sha: str, default_branch: str, is_default_head: bool) -> None:
@@ -379,6 +380,11 @@ def post_compute_guard(repo: str, sha: str, default_branch: str, is_default_head
     _post_guard("compute_guard", GUARD_CTX, repo, sha, default_branch, is_default_head)
 
 
+def post_secret_guard(repo: str, sha: str, default_branch: str, is_default_head: bool) -> None:
+    """No live credential in a bridged repo (leak hunt 09-24). Findings carry sha256[:8] only."""
+    _post_guard("secret_guard", SECRET_CTX, repo, sha, default_branch, is_default_head)
+
+
 def post_ci_hygiene(repo: str, sha: str, default_branch: str, is_default_head: bool) -> None:
     """House rule 6: lockfile-only installs, pinned images, no host-port services (warn-only)."""
     _post_guard("ci_hygiene", HYGIENE_CTX, repo, sha, default_branch, is_default_head)
@@ -438,6 +444,7 @@ def main() -> int:
                 post_statuses(repo, sha)
                 post_compute_guard(repo, sha, default_branch, sha == default_head)
                 post_ci_hygiene(repo, sha, default_branch, sha == default_head)
+                post_secret_guard(repo, sha, default_branch, sha == default_head)
         except Exception as e:  # one repo's failure must not hide the others'
             print(f"  FAILED {repo}: {e}")
             failed = 1

scripts/secret_guard.py

@@ -0,0 +1,110 @@
+#!/usr/bin/env python3
+"""Secret guard: no live credential lands in a bridged repo (leak hunt 09-24).
+
+Same walker, cache and GitHub posting as compute_guard / ci_hygiene
+(`windy-git/secret-guard`), but over EVERY text file, and a finding carries only
+"<kind> #<sha256[:8]>", never the value (house rule 10). Known fakes are allowed
+BY HASH in ci/secret-guard-allow.yml (repo + hashes + reason).
+
+  sudo python3 scripts/secret_guard.py report [repo ...]
+"""
+
+from __future__ import annotations
+
+import hashlib
+import os
+import re
+import sys
+from pathlib import Path
+
+import yaml
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+import compute_guard as cg  # noqa: E402  (shared walker, cache)
+import secret_shapes as ss  # noqa: E402
+
+ROOT = Path(__file__).resolve().parents[1]
+ALLOW_FILE = Path(os.environ.get("SECRET_GUARD_ALLOW", ROOT / "ci" / "secret-guard-allow.yml"))
+MODE = os.environ.get("SECRET_GUARD_MODE", "warn")
+NEVER = re.compile(r"(^|/)(node_modules|vendor|third_party)/")
+
+
+def path_ok(path: str) -> bool:
+    return not NEVER.search(path)
+
+
+def load_allow(path: Path = ALLOW_FILE) -> dict[str, set[str]]:
+    """{repo: {hash8, ...}}; every entry needs a reason."""
+    data = yaml.safe_load(path.read_text()) if path.exists() else {}
+    out: dict[str, set[str]] = {}
+    for e in (data or {}).get("allow") or []:
+        if not (e.get("repo") and e.get("hashes") and str(e.get("reason", "")).strip()):
+            raise ValueError(f"allow entry needs repo, hashes and a reason: {e}")
+        out.setdefault(e["repo"], set()).update(str(h) for h in e["hashes"])
+    return out
+
+
+def scan_line(path: str, text: str) -> list[tuple[str, str]]:
+    return [(kind, f"{kind} #{h}") for kind, h in ss.find(text)]
+
+
+def _drop_allowed(repo: str, findings, allow: dict[str, set[str]]):
+    ok = allow.get(repo, set())
+    return [f for f in findings if f.match.rsplit("#", 1)[-1] not in ok]
+
+
+def check(repo: str, sha: str, default_branch: str, is_default_head: bool):
+    bare = cg.WORK / f"{repo}.git"
+    if not bare.is_dir() or not cg.fetched(bare, sha):  # pushed after the fetch: next cycle
+        return None
+    allow = load_allow()
+    rules = hashlib.sha256(("|".join(rx.pattern for _, rx in ss.PATTERNS) + ss.PREFILTER).encode()).hexdigest()[:8]
+    kw = dict(line_fn=scan_line, path_ok=path_ok)
+    if is_default_head:
+        fs = cg.cached_scan(f"sec-tree:{repo}:{sha}:{rules}",
+                            lambda: cg.scan_tree(repo, bare, sha, [], prefilter=ss.PREFILTER, **kw))
+    else:
+        fs = cg.cached_scan(f"sec-pr:{repo}:{sha}:{rules}",
+                            lambda: cg.scan_added(repo, bare, f"refs/heads/{default_branch}", sha, [], **kw))
+    return _drop_allowed(repo, fs, allow)
+
+
+def status_for(findings, whole_tree: bool, grant=()):
+    """Same contract as the other guards. `grant` findings never block."""
+    scope = "in tree" if whole_tree else "added"
+    if not findings and grant:
+        g, n = grant[0], len(grant)
+        return "success", f"⚠ WARN (Grant-owned, not blocking): {n} secret-shaped string{'s' if n > 1 else ''} {scope}, e.g. {g.path}:{g.line} {g.match}"[:140], g
+    if not findings:
+        return "success", f"OK: no secret-shaped strings {scope}", None
+    f, n = findings[0], len(findings)
+    state = "failure" if MODE == "block" else "success"
+    lead = "BLOCKED" if MODE == "block" else "⚠ WARN (not blocking)"
+    return state, f"{lead}: {n} secret-shaped string{'s' if n > 1 else ''} {scope}, e.g. {f.path}:{f.line} {f.match}"[:140], f
+
+
+def report(repos: list[str]) -> int:
+    allow = load_allow()
+    total = 0
+    for repo in repos:
+        bare = cg.WORK / f"{repo}.git"
+        if not bare.is_dir():
+            continue
+        head = cg._git(bare, "symbolic-ref", "--short", "HEAD").strip()
+        sha = cg._git(bare, "rev-parse", head).strip()
+        fs = _drop_allowed(repo, cg.scan_tree(repo, bare, sha, [], line_fn=scan_line, path_ok=path_ok,
+                                               prefilter=ss.PREFILTER), allow)
+        total += len(fs)
+        print(f"## {repo} ({head} {sha[:7]}): {len(fs)} finding(s)")
+        for f in fs:
+            print(f"  {f.path}:{f.line}  {f.match}")
+    print(f"TOTAL {total}")
+    return 0
+
+
+if __name__ == "__main__":
+    if len(sys.argv) >= 2 and sys.argv[1] == "report":
+        default = os.environ.get("BRIDGE_REPOS", "").split() or sorted(
+            p.name.removesuffix(".git") for p in cg.WORK.glob("*.git"))
+        sys.exit(report(sys.argv[2:] or default))
+    sys.exit(__doc__)

scripts/secret_shapes.py

@@ -0,0 +1,40 @@
+"""Secret-shaped strings, shared by secret_guard (bridged repos) and
+public_secret_scan (weekly, every public repo). A finding NEVER carries the value:
+only its kind and sha256[:8] (house rule 10). Leak hunt 09-24: @Windy_0_bot's
+token sat in a public repo's test fixture for five months."""
+
+from __future__ import annotations
+
+import hashlib
+import re
+
+# (kind, regex). Order matters only for readability; each match is reported once.
+PATTERNS: list[tuple[str, re.Pattern[str]]] = [
+    ("telegram bot token", re.compile(r"(?<![0-9])[0-9]{8,10}:[A-Za-z0-9_-]{35}(?![A-Za-z0-9_-])")),
+    ("github token", re.compile(r"\b(?:gh[pousr]_[A-Za-z0-9]{36}|github_pat_[A-Za-z0-9_]{82})\b")),
+    ("aws access key", re.compile(r"\b(?:AKIA|ASIA)[0-9A-Z]{16}\b")),
+    ("slack token", re.compile(r"\bxox[abprs]-[A-Za-z0-9-]{10,}")),
+    ("anthropic key", re.compile(r"\bsk-ant-[A-Za-z0-9_-]{20,}")),
+    ("openai key", re.compile(r"\bsk-(?:proj-|svcacct-)?(?!ant-)[A-Za-z0-9_-]{32,}")),
+    ("stripe live key", re.compile(r"\b[rs]k_live_[A-Za-z0-9]{20,}")),
+    ("google api key", re.compile(r"\bAIza[0-9A-Za-z_-]{35}(?![0-9A-Za-z_-])")),
+    ("private key block", re.compile(r"-----BEGIN (?:RSA |EC |DSA |OPENSSH |ENCRYPTED )?PRIVATE KEY-----")),
+]
+
+# git grep -E (POSIX ERE) prefilter: cheap superset of PATTERNS.
+PREFILTER = ("[0-9]{8,10}:[A-Za-z0-9_-]{35}|gh[pousr]_[A-Za-z0-9]{36}|github_pat_|(AKIA|ASIA)[0-9A-Z]{16}"
+             "|xox[abprs]-|sk-ant-|sk-[A-Za-z0-9_-]{32}|sk-proj-|[rs]k_live_|AIza[0-9A-Za-z_-]{35}"
+             "|-----BEGIN [A-Z ]*PRIVATE KEY-----")
+
+
+def h8(value: str | bytes) -> str:
+    return hashlib.sha256(value.encode() if isinstance(value, str) else value).hexdigest()[:8]
+
+
+def find(text: str) -> list[tuple[str, str]]:
+    """[(kind, hash8)] for every secret-shaped string in `text`. Values never leave."""
+    out = []
+    for kind, rx in PATTERNS:
+        for m in rx.finditer(text):
+            out.append((kind, h8(m.group(0))))
+    return out