32 Commits

Author SHA1 Message Date
Kit OC5
1da4d39c0b bridge + sync: onboard windy-text, windy-call, windy-cell (telephony in scope, GitHub Actions dead since 08-14; deploy.yml disabled)
All checks were successful
check / gate (push) Successful in 14s
canary / probe (push) Successful in 6s
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-01 01:00:48 -04:00
Kit OC5
53fbcfe78f secret-guard allow: windy-code VS Code public aiKey, windytalk redaction fixture (hash not in lockbox)
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-01 00:59:59 -04:00
Kit OC5
1c552e94de secret-guard: Twilio shapes (SID/API key, 32-hex secret assignment) + per-kind warn mode
Windy Text 10-01: a live Twilio auth token sat in bridged-repo tests. Auth tokens are bare
32-hex, so they are matched only when assigned to a name containing token/secret/key/password;
hash is of the value alone. SECRET_GUARD_WARN_KINDS lets a new shape warn before it blocks.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-01 00:59:05 -04:00
Kit OC5
8690c4f8d3 secret-guard allow: 5 windy-agent #395 FAKE fixtures (weekly scan 09-28; verified never in the lockbox)
All checks were successful
check / gate (push) Successful in 17s
canary / probe (push) Successful in 5s
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-28 06:13:22 -04:00
Kit OC5
313f41b49c deploy: secret-guard BLOCK drop-in (orchestrator 09-24)
All checks were successful
check / gate (push) Successful in 8s
canary / probe (push) Successful in 6s
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-24 03:08:11 -04:00
Kit OC5
ea02ade0cb weekly public secret scan (all 5 GitHub accounts, full history, hash-only)
All checks were successful
check / gate (push) Successful in 13s
canary / probe (push) Successful in 4s
scripts/public_secret_scan.py: every PUBLIC repo, every object (incl.
unreachable + PR refs), secret_shapes patterns, excused by the secret-guard
allow list. Output JSON with hash8 + location only.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-24 03:05:16 -04:00
Kit OC5
497222094f secret-guard: triaged allow list (fakes by hash, test PEMs by path)
Private-key matches are only the BEGIN line (same hash everywhere), so they
are allowed by path+kind; everything else by hash. Real revoked tokens
(1354fc9b, d49dc2ba) are pinned by a test to never be allowed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-24 03:04:13 -04:00
Kit OC5
d28da26000 secret-guard: no live credentials in bridged repos (hash-only findings)
All checks were successful
check / gate (push) Successful in 22s
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>
2026-09-24 03:01:59 -04:00
Kit OC5
04c857654a rerun_ci.sh: find wg-q in the invoking user home under sudo
All checks were successful
check / gate (push) Successful in 10s
canary / probe (push) Successful in 5s
Under sudo ~ is /root, so the final run listing failed (windy-inbox #14).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-24 02:38:25 -04:00
Kit OC5
32512a32fc bridge windy-inbox (new private repo; Windy Drops is build lead)
All checks were successful
check / gate (push) Successful in 17s
canary / probe (push) Successful in 4s
sync REPOS + BRIDGE_REPOS + guards page (owner Windy Drops). Imported
writable into Gitea first (repo 164).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 20:51:09 -04:00
Kit OC5
b52e6b4784 bridge: post no-Docker smoke jobs (name regex skipped "no Docker")
All checks were successful
check / gate (push) Successful in 9s
canary / probe (push) Successful in 5s
windy-search #96 "Boot smoke (no Docker)" passed but was hidden by the
image-build name filter. Negative lookbehind for no/no-/without.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 20:11:55 -04:00
Kit OC5
6440c7d5f4 deploy: compute-guard BLOCK drop-in (live on Veron since 09-24 00:0xZ)
All checks were successful
check / gate (push) Successful in 12s
canary / probe (push) Successful in 7s
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 20:05:34 -04:00
Kit OC5
17acae6af6 guards: Grant-owned findings never block (compute-guard + ci-hygiene)
status_for(lane, whole_tree, grant=...): only lane-owned findings fail in
MODE=block; Grant-owned (ci/grant-owned.yml) post WARN. The bridge splits via
guards_report.split_grant; if the split cannot run it WARNs (never blocks).
Orchestrator 09-23: block compute-guard for lane-owned paths only.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 20:04:01 -04:00
Kit OC5
790d794900 bridge: drop eternitas ci/build from BRIDGE_NO_DAEMON (converted to ci/smoke, #179)
All checks were successful
check / gate (push) Successful in 10s
canary / probe (push) Successful in 6s
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 19:51:26 -04:00
Kit OC5
1b8cebbfe1 branding: friendly invite-only page instead of Gitea's raw Registration is disabled
A Windy account with no forge account (registration CLOSED at launch, Grant
09-23) lands on /user/link_account. Override shows invite-only + signed-in name
+ link to the dashboard; other cases = Gitea 1.24.6 template verbatim. No change
to who can register. Orchestrator ask 09-23.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 19:49:34 -04:00
Kit OC5
83fbf1f992 guards_report: chat lane is now the Windy Chat session
All checks were successful
check / gate (push) Successful in 10s
canary / probe (push) Successful in 6s
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 19:12:40 -04:00
Kit OC5
9b5334fee7 test: allow-list entries must be line-scoped (replaces is-empty check)
f71a5aa pushed with this test red (tail hid pytest rc); fixed here.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 19:11:53 -04:00
Kit OC5
f71a5aab39 ci-hygiene allow: windy-pro disabled deploy job (needs-docker false positive)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 19:11:20 -04:00
Kit OC5
f219437282 ci-hygiene report: same disabled-workflow filter as check()
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 19:10:35 -04:00
Kit OC5
b15584d33a ci-hygiene: needs-docker skips workflows disabled on Windy Git
deploy/release workflows run on the target host (real daemon) and are
disabled here (repo_unit DisabledWorkflows); one bounded query per process,
flag everything if it fails.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 19:10:11 -04:00
Kit OC5
0a57d96f2e bridge + ci-hygiene: Docker-needing CI jobs (option A)
- bridge: BRIDGE_NO_DAEMON names image-build jobs whose name lacks docker
  (default eternitas:ci/build); never posted, like the docker-named ones.
- ci-hygiene: flag docker build/buildx/run/compose, docker-compose and
  docker/build-push-action in workflow steps ("needs docker") with the fix:
  job services: + a no-Docker smoke test; the image builds at deploy.
- test_guards_report: owner column (14ed23a broke it).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 19:09:09 -04:00
Kit OC5
8b1ae4ac60 branding: home page says invite only (forge registration closed at launch)
Some checks failed
check / gate (push) Failing after 7s
canary / probe (push) Successful in 5s
Site-honesty re-audit 3.17 (Traveler): the home page promised sign-in with any
Windy account while ENABLE_AUTO_REGISTRATION=false.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 19:02:55 -04:00
Kit OC5
14ed23a9fe guards_report: owner-lane column (session names: 8c -> Windy Hub, calendar -> Windy Calender)
Some checks failed
check / gate (push) Failing after 10s
canary / probe (push) Failing after 1m5s
Orchestrator routing 09-23: hub/account-server/dashboard/site -> Windy Hub;
windy-calendar only -> Windy Calender; telemetry -> Windy Admin.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 18:44:11 -04:00
Kit OC5
c4799dcc13 ci: mount windy-pro's read-only build inputs into dind; allow exactly that path for jobs
All checks were successful
check / gate (push) Successful in 10s
canary / probe (push) Successful in 5s
Non-secret inputs git-ignored in windy-pro (models, linux-x64 portable
bundle, enter-monitor build) that build-desktop needs. Mounted :ro into
dind; valid_volumes allows only /ci-inputs/windy-pro; refresh-ci-inputs.sh
copies them from the frozen release clone (read-only on the source).
Invariant I-5 narrowed, not dropped: exactly that one path, read-only in
dind, no other service mounts it, still no docker socket (proven to fail
on :rw). Orchestrator-approved (option a). Applied in an idle window.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 17:28:48 -04:00
Kit OC5
683c01ec7d rerun_ci: Gitea stores repos lowercased on disk (WindyCloud failed)
All checks were successful
check / gate (push) Successful in 13s
canary / probe (push) Successful in 5s
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 17:22:17 -04:00
Kit OC5
4bb4893131 bridge: replace the mirror PR when a GitHub PR is retargeted to another base
All checks were successful
check / gate (push) Successful in 8s
canary / probe (push) Successful in 5s
The mirror kept its creation-time base forever. eternitas #167 was stacked
on fix/one-hallway, retargeted to main after #166 merged; ci.yml
(pull_request: branches [main]) then silently never ran for it, while
unfiltered workflows did. An edited event triggers nothing, so close the
stale mirror and open a fresh one on the new base (runs CI at once).
An unknown base is left alone, never guessed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 16:16:15 -04:00
Kit OC5
9e9eb08d96 compute guard: line-scoped allow entries; allow MindKeychain.jsx's two OAuth endpoints only
All checks were successful
check / gate (push) Successful in 12s
canary / probe (push) Successful in 4s
Allow entries may carry matches: (regexes); then only matching lines are
allowed, so an allowed file can't smuggle in a new call. windy-pro #609
MindKeychain.jsx: openrouter.ai/auth? and /api/v1/auth/keys (BYOK key
acquisition via OAuth PKCE, no inference; successor of the MindPanel
allow, ADR-064). An inference call in the same file still flags (tested).
Orchestrator-approved.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 15:46:53 -04:00
Kit OC5
dece301bf0 grant-owned: windy-pro ROOT .env.example (desktop BYOK dev fallback; hub reads neither)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 15:45:33 -04:00
Kit OC5
87dcddb87d guards_report: one live status page for compute-guard + ci-hygiene
All checks were successful
check / gate (push) Successful in 15s
Scans every bridged default branch with both guards; lane-owned vs
Grant-owned (ci/grant-owned.yml: windy-pro desktop paths + its desktop CI
jobs, attributed per job) so Grant's code never holds up a block.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 15:38:21 -04:00
Kit OC5
a5f7b036a8 guards: skip a commit the sync hasn't fetched yet, quietly (race, not error)
All checks were successful
check / gate (push) Successful in 15s
canary / probe (push) Successful in 4s
The bridge reads PR/default heads from GitHub after the sync's fetch; a
push in between isn't in the clone until the next cycle. Both guards logged
a CalledProcessError for it (windy-pro main 40 s after the fetch). Now
None = nothing posted this cycle; the next one scans it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 15:36:19 -04:00
Kit OC5
255aa58b35 ci-hygiene guard: lockfile-only installs, pinned images, no host-port services (warn-only)
All checks were successful
check / gate (push) Successful in 19s
House rule 6 (09-23). The bridge now also posts windy-git/ci-hygiene on
every PR head (added lines) and default branch (whole files), scanning CI
workflows and Dockerfiles for: floating pip / uv pip installs (not -r,
not --no-deps, not exact pins), uv sync without --locked/--frozen,
npm install instead of npm ci (unless every package is exact-pinned),
yarn/pnpm without a frozen lockfile, :latest images and COPY lock* globs
(Windy Mail #147), and CI services publishing a HOST port (every job
shares one dind: Windy Mind runs 147/176 died on 5432). Warn-only;
CI_HYGIENE_MODE=block later. Allow-list ci/ci-hygiene-allow.yml (empty).

compute_guard's walker is now parameterised (line_fn / path_ok /
prefilter) so both guards share one scanner, cache and allow loader; the
bridge posts both through one _post_guard. Today: 95 issues in 21 repos;
windy-git, calendar, traveler, traveler-site clean.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 15:30:51 -04:00
Kit OC5
e64a1b5fcb deps: install from uv.lock (hash-pinned) in the image and CI, never floating
All checks were successful
check / gate (push) Successful in 10s
canary / probe (push) Successful in 5s
The API image did 'pip install -e .' and CI 'pip install -e ".[dev]"':
every rebuild could ship newer fastapi/starlette/pydantic than CI tested
(Windy Cloud hit exactly this 09-23). uv.lock is cut to EXACTLY what prod
runs now (42 runtime pkgs, 0 differences; botocore/pyjwt held back to
prod's versions). Image: uv 0.12.5 exports, pip --require-hashes installs,
same layout. CI: uv sync --locked (also fails on a stale lock).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-23 15:25:18 -04:00
26 changed files with 3039 additions and 48 deletions

View File

@@ -51,9 +51,10 @@ jobs:
- name: install
run: |
python3 --version
python3 -m venv .venv
.venv/bin/pip install -q --upgrade pip
.venv/bin/pip install -e ".[dev]"
# From uv.lock, never floating: CI tests exactly what the image ships.
# --locked also FAILS if pyproject.toml changed without re-locking.
python3 -m pip install -q uv==0.12.5
uv sync --locked --extra dev
- name: lint
run: .venv/bin/ruff check api scripts

View File

@@ -10,10 +10,19 @@ WORKDIR /app
RUN apt-get update && apt-get install -y --no-install-recommends git curl \
&& rm -rf /var/lib/apt/lists/*
COPY pyproject.toml ./
RUN pip install --no-cache-dir -e .
# Dependencies come from uv.lock, hash-pinned, never "latest at build time".
# Floating installs meant a rebuild could ship different fastapi/starlette/
# pydantic than CI tested (Windy Cloud's OpenAPI drift, 09-23). The lock was
# cut to exactly what prod ran then. uv only exports; pip installs, so the
# image layout (system python, uvicorn on PATH) is unchanged.
COPY --from=ghcr.io/astral-sh/uv:0.12.5 /uv /usr/local/bin/uv
COPY pyproject.toml uv.lock ./
RUN uv export --frozen --no-dev --no-emit-project -o /tmp/requirements.txt \
&& pip install --no-cache-dir --require-hashes -r /tmp/requirements.txt \
&& rm /tmp/requirements.txt
COPY api ./api
RUN pip install --no-cache-dir --no-deps -e .
COPY alembic ./alembic
COPY alembic.ini ./
COPY scripts ./scripts

View File

@@ -0,0 +1,164 @@
"""CI hygiene guard (house rule 6): lockfile-only installs, no host-port services."""
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("ci_hygiene", ROOT / "scripts" / "ci_hygiene.py")
hy = importlib.util.module_from_spec(_spec)
sys.modules["ci_hygiene"] = hy
_spec.loader.exec_module(hy)
WF = ".github/workflows/ci.yml"
@pytest.mark.parametrize("path, text", [
(WF, " .venv/bin/pip install -e \".[dev]\""), # windy-git's own, before e64a1b5
("Dockerfile", "RUN pip install --no-cache-dir -e ."), # windy-git image, before e64a1b5
(WF, " - run: uv pip install -e \".[dev]\""), # WindyCloud #109's CI
(WF, " run: pip install fastapi uvicorn"),
(WF, " - run: uv sync --all-extras"), # windy-mind style, not locked
(WF, " - run: npm install"), # windy-drops / windytalk
(WF, " - run: npm install --no-save --no-audit --no-fund jsdom"), # windy-pro reality-check
(WF, " - run: yarn install"),
(WF, " - run: cd web && pnpm install"),
("docker/api.Dockerfile", "RUN apt-get update && pip install requests"),
])
def test_floating_installs_are_flagged(path, text):
assert [k for k, _ in hy.scan_line(path, text)] == ["floating install"]
@pytest.mark.parametrize("path, text", [
(WF, " python3 -m pip install -q uv==0.12.5"), # exact tool pin
(WF, " uv sync --locked --extra dev"),
(WF, " - run: uv sync --frozen"),
(WF, " - run: npm ci"),
(WF, " - run: npm install --no-save jsdom@24.1.0"),
(WF, " - run: pip install -r requirements.lock --require-hashes"),
(WF, " - run: pip install -r requirements.txt"),
("Dockerfile", " && pip install --no-cache-dir --require-hashes -r /tmp/requirements.txt \\\\"),
("Dockerfile", "RUN pip install --no-cache-dir --no-deps -e ."), # project only, deps from the lock
(WF, " .venv/bin/pip install -q --upgrade pip"),
(WF, " - run: yarn install --frozen-lockfile"),
(WF, " # - run: npm install (commented out)"),
(WF, " - run: echo 'pip is great'"),
])
def test_locked_or_pinned_installs_pass(path, text):
assert hy.scan_line(path, text) == []
@pytest.mark.parametrize("text, port", [
(" - 5432:5432", "5432"), # windy-mind / eternitas (collided 09-23)
(" - '15432:5432'", "15432"), # WindyCloud
(' - "6379:6379"', "6379"),
])
def test_services_publishing_a_host_port_are_flagged(text, port):
[(kind, match)] = hy.scan_line(WF, text)
assert kind == "host port" and port in match
def test_host_port_rule_is_for_workflows_only():
assert hy.scan_line("docker-compose.yml", " - 5432:5432") == []
@pytest.mark.parametrize("path, ok", [
(".github/workflows/ci.yml", True), (".gitea/workflows/check.yaml", True),
("Dockerfile", True), ("api/Dockerfile.prod", True), ("docker/web.Dockerfile", True),
("scripts/setup.sh", False), ("README.md", False), ("node_modules/x/Dockerfile", False),
(".github/lint/x.yml", False),
])
def test_scope_is_ci_workflows_and_dockerfiles(path, ok):
assert hy.path_ok(path) is ok
def test_warn_mode_never_turns_red(monkeypatch):
monkeypatch.setattr(hy, "MODE", "warn")
state, desc, f = hy.status_for([hy.cg.Finding(WF, 12, "floating install", "npm install (use npm ci)")], True)
assert state == "success" and desc.startswith("⚠ WARN (not blocking): 1 CI hygiene issue in CI/Dockerfiles")
def test_allow_file_is_line_scoped_exceptions_only():
"""Every exception is line-scoped (`matches`), so an allowed file can't hide a
NEW floating install or docker step. Today: windy-pro's if:false deploy job."""
allow = hy.cg.load_allow(hy.ALLOW_FILE)
assert [(e["repo"], e["paths"]) for e in allow] == [("windy-pro", [".github/workflows/ci.yml"])]
assert all(e.get("matches") for e in allow)
ok = " run: docker build -f account-server/Dockerfile -t windy-pro:${{ github.sha }} ."
new = " run: docker build -t windy-pro-api ."
assert hy.cg.allowed("windy-pro", WF, allow, ok)
assert not hy.cg.allowed("windy-pro", WF, allow, new)
assert not hy.cg.allowed("windy-chat", WF, allow, ok)
@pytest.mark.parametrize("path, text, want", [
("Dockerfile", "COPY --from=ghcr.io/astral-sh/uv:latest /uv /usr/local/bin/uv", "ghcr.io/astral-sh/uv:latest"), # Mail #147
("Dockerfile", "FROM python:latest", "python:latest"),
("Dockerfile", "FROM --platform=linux/amd64 node:latest AS web", "node:latest"),
(WF, " image: postgres:latest", "postgres:latest"),
(WF, " - uses: docker://ghcr.io/foo/bar:latest", "ghcr.io/foo/bar:latest"),
])
def test_latest_images_are_flagged(path, text, want):
hits = hy.scan_line(path, text)
assert ("floating image", want) in hits
@pytest.mark.parametrize("text", [
"COPY pyproject.toml uv.lock* ./", # Windy Mail #147
"COPY package.json package-lock.json* ./",
])
def test_optional_lock_globs_are_flagged(text):
assert [k for k, _ in hy.scan_line("Dockerfile", text)] == ["optional lock"]
@pytest.mark.parametrize("path, text", [
("Dockerfile", "COPY --from=ghcr.io/astral-sh/uv:0.12.5 /uv /usr/local/bin/uv"),
("Dockerfile", "FROM python:3.12-slim"),
("Dockerfile", "COPY pyproject.toml uv.lock ./"),
("Dockerfile", "COPY src/*.py ./src/"),
("Dockerfile", "RUN echo latest release notes"),
])
def test_pinned_images_and_real_locks_pass(path, text):
assert hy.scan_line(path, text) == []
@pytest.mark.parametrize("text", [
" - run: docker compose -f docker-compose.yml -f docker-compose.ci.yml build", # eternitas ci/build
" run: docker build -t windy-mail .",
" - run: docker-compose up -d",
" run: docker buildx build --load .",
" - uses: docker/build-push-action@v6",
])
def test_docker_in_ci_is_flagged_with_the_fix(text):
hits = hy.scan_line(WF, text)
assert [k for k, _ in hits] == ["needs docker"]
assert "no-Docker smoke test" in hits[0][1]
@pytest.mark.parametrize("path, text", [
("Dockerfile", "RUN docker build ."), # not a workflow
(WF, " run: ssh host 'docker compose up -d'"), # remote host has a daemon
(WF, " # docker compose build"), # comment
(WF, " run: echo docker build"),
])
def test_docker_not_flagged_outside_ci_steps(path, text):
assert [k for k, _ in hy.scan_line(path, text) if k == "needs docker"] == []
def test_needs_docker_skips_workflows_disabled_on_windy_git(monkeypatch):
"""deploy.yml runs on the target host (a real daemon); Gitea has it disabled here."""
F = hy.cg.Finding
monkeypatch.setattr(hy, "_DISABLED", {"eternitas": {"deploy.yml"}})
got = hy._runs_here("Eternitas", [
F(".github/workflows/deploy.yml", 70, "needs docker", "docker compose in CI"),
F(".github/workflows/ci.yml", 176, "needs docker", "docker compose in CI"),
F(".github/workflows/deploy.yml", 12, "floating install", "npm install"),
])
assert [(f.path.rsplit("/", 1)[1], f.kind) for f in got] == [
("ci.yml", "needs docker"), ("deploy.yml", "floating install")]
assert hy._runs_here("eternitas", None) is None

View File

@@ -182,3 +182,50 @@ def test_code_with_a_trailing_comment_still_counts():
def test_windy_pro_desktop_is_byok_but_the_account_server_is_not():
assert cg.allowed("windy-pro", "src/client/desktop/main.js", ALLOW)
assert not cg.allowed("windy-pro", "account-server/src/routes/translations.ts", ALLOW)
def test_a_commit_not_fetched_yet_is_skipped_not_an_error(tmp_path, monkeypatch):
bare, sha = _repo(tmp_path, {"app/llm.py": "import anthropic\n"})
monkeypatch.setattr(cg, "WORK", tmp_path)
monkeypatch.setattr(cg, "CACHE", tmp_path / "cache.json")
(tmp_path / "windy-chat.git").symlink_to(bare)
assert cg.check("windy-chat", "f" * 40, "main", True) is None # pushed after the fetch
assert [f.kind for f in cg.check("windy-chat", sha, "main", True)] == ["provider SDK"]
KEYCHAIN = "src/client/web/src/pages/panels/MindKeychain.jsx"
def test_scoped_allow_admits_only_the_oauth_endpoints():
ok = [
" window.location.href = `https://openrouter.ai/auth?callback_url=${encodeURIComponent(callback)}`",
" const res = await fetch('https://openrouter.ai/api/v1/auth/keys', {",
]
for line in ok:
assert cg.allowed("windy-pro", KEYCHAIN, ALLOW, line)
# an inference call smuggled into the same file still flags
assert not cg.allowed("windy-pro", KEYCHAIN, ALLOW,
" await fetch('https://openrouter.ai/api/v1/chat/completions', {")
# a scoped entry never allows a line it can't see
assert not cg.allowed("windy-pro", KEYCHAIN, ALLOW)
def test_scoped_allow_in_a_real_diff():
diff = f"""--- /dev/null
+++ b/{KEYCHAIN}
@@ -0,0 +1,3 @@
+ window.location.href = `https://openrouter.ai/auth?callback_url=x`
+ const res = await fetch('https://openrouter.ai/api/v1/auth/keys', {{
+ await fetch('https://openrouter.ai/api/v1/chat/completions', {{
"""
fs = cg.parse_added("windy-pro", diff, ALLOW)
assert [(f.line, f.match) for f in fs] == [(3, "openrouter.ai")]
def test_block_mode_never_blocks_grant_owned(monkeypatch):
monkeypatch.setattr(cg, "MODE", "block")
g = cg.Finding("src/client/desktop/x.js", 9, "provider host", "api.openai.com")
state, desc, f = cg.status_for([], whole_tree=True, grant=[g])
assert state == "success" and desc.startswith("⚠ WARN (Grant-owned, not blocking): 1") and f is g
lane = cg.Finding("a.py", 3, "provider host", "x")
assert cg.status_for([lane], whole_tree=True, grant=[g])[0] == "failure"

View File

@@ -0,0 +1,77 @@
"""guards_report: job attribution and the Grant-owned split."""
from __future__ import annotations
import importlib.util
import sys
from pathlib import Path
import yaml
ROOT = Path(__file__).resolve().parents[2]
sys.path.insert(0, str(ROOT / "scripts"))
_spec = importlib.util.spec_from_file_location("guards_report", ROOT / "scripts" / "guards_report.py")
gr = importlib.util.module_from_spec(_spec)
sys.modules["guards_report"] = gr
_spec.loader.exec_module(gr)
OWNED = yaml.safe_load((ROOT / "ci" / "grant-owned.yml").read_text())["grant_owned"]
WF = """name: CI
on:
push:
jobs:
reality-check:
runs-on: x
steps:
- run: npm install jsdom
test-backend:
runs-on: x
steps:
- run: pip install pytest
"""
def test_job_of_attributes_lines_to_their_job():
assert gr.job_of(WF, 3) is None # `on:` block, not a job
assert gr.job_of(WF, 8) == "reality-check"
assert gr.job_of(WF, 12) == "test-backend"
def test_windy_pro_desktop_jobs_and_paths_are_grant_owned():
ci = ".github/workflows/ci.yml"
assert gr.grant_owned("windy-pro", ci, "reality-check", OWNED)
assert gr.grant_owned("windy-pro", ci, "build-electron", OWNED)
assert not gr.grant_owned("windy-pro", ci, "test-backend", OWNED) # server side: 8c
assert gr.grant_owned("windy-pro", ".github/workflows/release-mac.yml", None, OWNED)
assert gr.grant_owned("windy-pro", "src/client/desktop/main.js", None, OWNED)
assert not gr.grant_owned("windy-pro", "services/account-server/Dockerfile", None, OWNED)
assert not gr.grant_owned("windy-chat", "src/client/desktop/main.js", None, OWNED)
def test_render_splits_lane_and_grant_counts():
F = gr.cg.Finding
res = {"windy-pro": {"sha": "a" * 40, "compute": [],
"hygiene": [(F("ci.yml", 8, "floating install", "npm install"), "reality-check", True),
(F("ci.yml", 12, "floating install", "pip x"), "test-backend", False)]},
"windy-git": {"sha": "b" * 40, "compute": [], "hygiene": []}}
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 | 0 | clean ✅ |" in md
assert "| windy-pro | Windy Hub | aaaaaaa |" in md # owner = session to message
assert "(job reality-check)" in md
def test_windy_pro_root_env_example_is_grant_owned_but_not_the_account_servers():
assert gr.grant_owned("windy-pro", ".env.example", None, OWNED)
assert not gr.grant_owned("windy-pro", "account-server/.env.example", None, OWNED)
def test_split_grant_sends_desktop_code_to_grant(monkeypatch):
F = gr.cg.Finding
fs = [F("src/client/desktop/main.js", 3, "provider host", "x"),
F("account-server/src/llm.ts", 5, "provider host", "y")]
lane, grant = gr.split_grant("windy-pro", "a" * 40, fs)
assert [f.path for f in grant] == ["src/client/desktop/main.js"]
assert [f.path for f in lane] == ["account-server/src/llm.ts"]
assert gr.split_grant("windy-chat", "a" * 40, fs) == (fs, [])

View File

@@ -351,10 +351,12 @@ class _Guard:
return self.findings
@staticmethod
def status_for(findings, whole_tree):
def status_for(findings, whole_tree, grant=()):
if not findings and grant:
return "success", f"GRANT-WARN {len(grant)}", grant[0]
if not findings:
return "success", "OK: clean", None
return "success", f"WARN {len(findings)}", findings[0]
return "failure", f"BLOCK {len(findings)}", findings[0]
class _F:
@@ -366,12 +368,12 @@ def test_guard_posts_warn_with_a_link_to_the_first_finding(fake, monkeypatch):
monkeypatch.setitem(sys.modules, "compute_guard", _Guard([_F()]))
bridge.post_compute_guard("windy-chat", SHA, "main", False)
assert [(p["context"], p["state"], p["description"]) for p in f.posted] == [
("windy-git/compute-guard", "success", "WARN 1")]
("windy-git/compute-guard", "failure", "BLOCK 1")]
assert f.posted[0]["target_url"].endswith(f"/src/commit/{SHA}/app/llm.py#L7")
def test_guard_same_status_is_not_reposted(fake, monkeypatch):
f = fake(statuses=[{"context": "windy-git/compute-guard", "state": "success", "description": "WARN 1"}])
f = fake(statuses=[{"context": "windy-git/compute-guard", "state": "failure", "description": "BLOCK 1"}])
monkeypatch.setitem(sys.modules, "compute_guard", _Guard([_F()]))
bridge.post_compute_guard("windy-chat", SHA, "main", False)
assert f.posted == []
@@ -382,3 +384,80 @@ def test_guard_that_cannot_run_posts_nothing(fake, monkeypatch):
monkeypatch.setitem(sys.modules, "compute_guard", _Guard(None))
bridge.post_compute_guard("windy-chat", SHA, "main", True)
assert f.posted == []
def test_ci_hygiene_posts_under_its_own_context(fake, monkeypatch):
f = fake(statuses=[{"context": "windy-git/compute-guard", "state": "failure", "description": "BLOCK 1"}])
monkeypatch.setitem(sys.modules, "ci_hygiene", _Guard([_F()]))
bridge.post_ci_hygiene("windy-chat", SHA, "main", True)
# the compute-guard status with the same description must not suppress it
assert [(p["context"], p["description"]) for p in f.posted] == [("windy-git/ci-hygiene", "BLOCK 1")]
def test_retargeted_pr_gets_a_fresh_mirror_on_the_new_base(fake):
# eternitas #167: stacked on fix/one-hallway, retargeted to main on GitHub.
gh = [{"number": 167, "title": "feat", "html_url": "u",
"head": {"ref": "feat/x", "sha": SHA, "repo": {"full_name": f"{bridge.GH_OWNER}/eternitas"}},
"base": {"ref": "main"}}]
wg = [{"number": 9, "title": "[GH#167] feat", "base": {"ref": "fix/one-hallway"}}]
f = fake(gh_prs=gh, wg_prs=wg)
assert bridge.sync_prs("eternitas") == [SHA]
assert f.closed == [f"/repos/{bridge.WG_OWNER}/eternitas/pulls/9"]
assert [(o["base"], o["head"]) for o in f.opened] == [("main", "feat/x")]
def test_unchanged_base_leaves_the_mirror_alone(fake):
gh = [{"number": 5, "title": "t", "html_url": "u",
"head": {"ref": "b", "sha": SHA, "repo": {"full_name": f"{bridge.GH_OWNER}/windy-chat"}},
"base": {"ref": "main"}}]
f = fake(gh_prs=gh, wg_prs=[{"number": 3, "title": "[GH#5] t", "base": {"ref": "main"}}])
bridge.sync_prs("windy-chat")
assert f.closed == [] and f.opened == []
def test_named_no_daemon_job_is_not_posted_for_that_repo_only(fake, monkeypatch):
"""eternitas ci/build needs Docker but its name doesn't say so (option A, 09-23)."""
monkeypatch.setattr(bridge, "NO_DAEMON_NAMED", {"eternitas": {"ci/build"}})
runs = [_run(1, "ci.yml", "build", "failure"), _run(2, "ci.yml", "test", "success")]
f = fake(runs=runs)
bridge.post_statuses("eternitas", SHA)
assert [p["context"] for p in f.posted] == ["windy-git/ci/test"]
f2 = fake(runs=runs)
bridge.post_statuses("windy-chat", SHA)
assert sorted(p["context"] for p in f2.posted) == ["windy-git/ci/build", "windy-git/ci/test"]
def test_default_no_daemon_named_is_empty():
"""eternitas converted its ci/build to a no-Docker ci/smoke (#179); nothing left."""
assert bridge.NO_DAEMON_NAMED == {}
def test_grant_owned_findings_never_block(fake, monkeypatch):
"""Orchestrator 09-23: compute-guard blocks lane-owned code only."""
f = fake()
monkeypatch.setitem(sys.modules, "compute_guard", _Guard([_F()]))
monkeypatch.setitem(sys.modules, "guards_report",
type("GR", (), {"split_grant": staticmethod(lambda r, s, fs: ([], list(fs)))}))
bridge.post_compute_guard("windy-pro", SHA, "main", True)
assert [(p["state"], p["description"]) for p in f.posted] == [("success", "GRANT-WARN 1")]
def test_failed_grant_split_warns_instead_of_blocking(fake, monkeypatch):
def boom(*a):
raise RuntimeError("no bare clone")
f = fake()
monkeypatch.setitem(sys.modules, "compute_guard", _Guard([_F()]))
monkeypatch.setitem(sys.modules, "guards_report", type("GR", (), {"split_grant": staticmethod(boom)}))
bridge.post_compute_guard("windy-pro", SHA, "main", True)
assert [p["state"] for p in f.posted] == ["success"]
@pytest.mark.parametrize("name, hidden", [
("Docker Build", True), ("docker-build", True), ("Docker build", True), ("docker", True),
("Boot smoke (no Docker)", False), ("smoke (no-docker)", False), ("boot without Docker", False),
("smoke", False),
])
def test_no_docker_smoke_jobs_are_posted(name, hidden):
"""windy-search #96: its replacement job says "no Docker" and was hidden."""
assert bridge.needs_daemon("windy-search", "ci", name) is hidden

View File

@@ -0,0 +1,148 @@
"""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")
allow = {"windy-chat": {"hashes": {ss.h8(TG)}, "paths": []}}
assert sg._drop_allowed("windy-chat", [fake, real], allow) == [real]
assert sg._drop_allowed("windy-mail", [fake], allow) == [fake]
def test_private_key_blocks_are_allowed_by_path_never_by_hash():
F = sg.cg.Finding
hdr = "private key block #" + ss.h8("-----BEGIN PRIVATE KEY-----")
test_key = F("tests/keys/test.pem", 1, "private key block", hdr)
prod_key = F("deploy/prod.pem", 1, "private key block", hdr)
allow = {"r": {"hashes": {hdr.rsplit("#", 1)[1]}, "paths": [("tests/keys/*", {"private key block"})]}}
assert sg._drop_allowed("r", [test_key, prod_key], allow) == [prod_key]
def test_path_allow_cannot_cover_real_token_kinds(tmp_path):
bad = tmp_path / "a.yml"
bad.write_text("allow:\n - repo: r\n paths: [tests/*]\n kinds: [telegram bot token]\n reason: no\n")
with pytest.raises(ValueError):
sg.load_allow(bad)
def test_shipped_allow_file_never_excuses_the_real_leaked_tokens():
a = sg.load_allow()
every = set().union(*(v["hashes"] for v in a.values())) if a else set()
assert not {"1354fc9b", "d49dc2ba"} & every # real (now revoked) credentials: remove, never allow
def test_allow_file_loads_and_needs_reasons(tmp_path):
assert 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")
def test_public_scan_excuses_by_hash_and_by_path():
spec = importlib.util.spec_from_file_location("public_secret_scan", ROOT / "scripts" / "public_secret_scan.py")
ps = importlib.util.module_from_spec(spec)
spec.loader.exec_module(ps)
allow = {"windy-agent": {"hashes": {"aaaa1111"}, "paths": [("tests/keys/*", {"private key block"})]}}
assert ps.excused("windy-agent", "openai key", "aaaa1111", ["tests/x.py"], allow)
assert not ps.excused("windy-agent", "telegram bot token", "1354fc9b", ["tests/test_log_redaction.py"], allow)
assert ps.excused("windy-agent", "private key block", "ffff0000", ["tests/keys/a.pem"], allow)
# a PEM header anywhere outside the allowed paths still counts
assert not ps.excused("windy-agent", "private key block", "ffff0000", ["tests/keys/a.pem", "deploy/k.pem"], allow)
assert not ps.excused("other", "openai key", "aaaa1111", ["x"], allow)
HEX32 = "0123456789abcdef" * 2 # synthetic
def test_twilio_shapes_hash_only_and_no_md5_noise():
kinds = lambda t: [k for k, _ in ss.find(t)] # noqa: E731
assert kinds(f'TWILIO_AUTH_TOKEN = "{HEX32}"') == ["32-hex secret assignment"]
assert kinds(f"auth_token: {HEX32}") == ["32-hex secret assignment"]
assert kinds("AC" + HEX32) == ["twilio sid/api key"]
assert kinds("SK" + HEX32) == ["twilio sid/api key"]
# plain md5 / uuid-without-dashes / a 64-hex sha256 are NOT secrets by shape
assert kinds(f"md5 = {HEX32}") == []
assert kinds(f"checksum_key = {HEX32}{HEX32}") == []
assert kinds(f"name = 'x{HEX32}'") == []
# the hash is of the value alone, so renaming the variable keeps the same allow hash
a = ss.find(f"A_TOKEN={HEX32}")[0][1]
b = ss.find(f"OTHER_SECRET: '{HEX32}'")[0][1]
assert a == b == ss.h8(HEX32)
assert HEX32 not in repr(ss.find(f"A_TOKEN={HEX32}"))
def test_warn_kinds_do_not_block(monkeypatch):
f = sg.cg.Finding("a.py", 1, "32-hex secret assignment", "32-hex secret assignment #abcd1234")
monkeypatch.setattr(sg, "MODE", "block")
monkeypatch.setattr(sg, "WARN_KINDS", {"32-hex secret assignment"})
assert sg.status_for([f], True)[0] == "success"
monkeypatch.setattr(sg, "WARN_KINDS", set())
assert sg.status_for([f], True)[0] == "failure"

13
ci/ci-hygiene-allow.yml Normal file
View File

@@ -0,0 +1,13 @@
# CI hygiene allow-list: installs that may float, or services that may publish
# a host port. House rule 6 (09-23): installs come from a lockfile. Every entry
# is an exception and MUST say why. Paths are fnmatch globs from the repo root.
# Owner: Windy Git lane (13); changes go through the orchestrator.
allow:
- repo: windy-pro
paths: [".github/workflows/ci.yml"]
# ONLY the old deploy job's two docker lines. That job is `if: false`
# (CD boundary, 2026-07), and the compose line runs ON windyword.ai inside
# the ssh string. Any other docker step in ci.yml still flags.
matches: ['docker build -f account-server/Dockerfile -t windy-pro:', 'docker compose down && docker compose up -d --build']
reason: "needs-docker false positive: the deploy job is if: false and its compose runs on the remote host over ssh. Added with the needs-docker rule (orchestrator option A, 09-23)."

View File

@@ -35,6 +35,13 @@ allow:
paths: ["src/client/web/src/pages/panels/MindPanel.jsx"]
reason: "Validates the USER's own OpenRouter key for BYOK (audit #10); spends no house money."
- repo: windy-pro
paths: ["src/client/web/src/pages/panels/MindKeychain.jsx"]
# ONLY these two endpoints: any other openrouter.ai call in this file (e.g.
# /api/v1/chat, i.e. inference) still flags. Orchestrator-approved 09-23.
matches: ['openrouter\.ai/auth\?', 'openrouter\.ai/api/v1/auth/keys']
reason: "BYOK key acquisition via OpenRouter OAuth PKCE; no inference; successor of MindPanel allow (ADR-064)."
- repo: windy-git
paths: ["scripts/compute_guard.py", "ci/compute-guard-allow.yml"]
reason: "The guard's own pattern list and this file."

19
ci/grant-owned.yml Normal file
View File

@@ -0,0 +1,19 @@
# Code Grant owns directly (orchestrator, 09-23): guard findings here are listed
# SEPARATELY in the guards status page and never hold up "block". Changes to
# these files are proposals for Grant / Windy Word 44, not a lane's fix.
grant_owned:
- repo: windy-pro
reason: "Windy Word desktop (Electron) + its release/installer builds: Grant's, built from the Mac mini."
paths:
- "src/client/desktop/*"
# ROOT .env.example only (exact path): documents the desktop app's BYOK dev
# fallback keys; the hub reads neither (8c, 09-23). account-server/.env.example
# is NOT matched and stays the hub lane's.
- ".env.example"
- "installer-v2/*"
- ".github/workflows/build-windows.yml"
- ".github/workflows/release-mac.yml"
- ".github/workflows/build-installer.yml"
- ".github/workflows/build-offline-installers.yml"
jobs:
".github/workflows/ci.yml": [reality-check, build-desktop, test-installer, build-electron]

59
ci/secret-guard-allow.yml Normal file
View File

@@ -0,0 +1,59 @@
# Secret guard allow-list: KNOWN FAKE values that look like secrets (test
# fixtures, docs). Allowed BY HASH (sha256[:8] of the value), so a real secret
# in the same file still flags. Private-key blocks (the match is only the BEGIN
# line, same hash everywhere) are allowed by PATH + kind instead.
# Every entry MUST say why. Owner: Windy Git lane (13); changes via the orchestrator.
# Triage 09-24 (values never printed): each hash checked against every version of
# the lockbox; fakes judged by impossible length for the kind (real Anthropic keys
# ~108 chars, OpenAI 51 or 160+), fake-words, or identity with upstream public
# fixtures. NOT allowed, remove instead: 1354fc9b (old @Windy_0_bot token) and
# d49dc2ba (old Anthropic key), both real and revoked, in public windy-agent.
allow:
- repo: windy-code
hashes: [ac9265e5, 46eb1235]
reason: "Upstream microsoft/vscode terminalEnvironment.test.ts fixtures (identical hash upstream; public)."
- repo: windy-code
paths: ["build/azure-pipelines/common/publish.ts"]
kinds: [private key block]
reason: "Upstream VS Code build script (PEM header string in code, not a key)."
- repo: windy-agent
hashes: [a9235a6d, dd6a2baa, 02c362d8, a6f6ff79, f7503b21, d0c94833, 4ab092e3, e3aa1eb8, 833382ee]
reason: "Redaction/sanitizer test fixtures; lengths impossible for real Anthropic/OpenAI keys; never in the lockbox."
- repo: windy-agent
hashes: [89bd408f, b8c94b72, bf23cdf5, d1d85dfe, e3e07f06]
reason: "09-24 #395 replacement fixtures (each contains FAKE; token-SHAPED on purpose so redaction tests prove real tokens are scrubbed); verified by Windy Agent sha-for-sha against every lockbox version: never real. Weekly scan 09-28."
- repo: windy-agent
paths: ["tests/test_agent_keys.py"]
kinds: [private key block]
reason: "Test-generated key material for agent-key tests."
- repo: windy-mind
hashes: [f8a630b2]
reason: "Provider test fixture (27 chars; a real Anthropic key is ~108)."
- repo: windy-pro
hashes: [756de8d8, 7828319d, 1a5d44a2]
reason: ".env.production.example placeholder + crash-summary test fixtures (AWS doc EXAMPLE key shape, short fake Slack token)."
- repo: windy-pro
paths: ["account-server/docs/oauth-providers.md"]
kinds: [private key block]
reason: "Docs show the PEM header format; no key material."
- repo: windytalk
hashes: [baf8656a]
reason: "Diagnostics redaction test fixture (fake-word in value)."
- repo: eternitas
paths: ["tests/golden_vectors/**", "tests/test_soul_vault_key_separation.py"]
kinds: [private key block]
reason: "Test vectors and throwaway keys for signature/vault tests."
- repo: windy-drops
paths: ["tools/conformance/test-keys/*"]
kinds: [private key block]
reason: "Conformance-suite test keys (named test-private.pem)."
- repo: windy-git
paths: ["api/tests/test_secret_guard.py"]
kinds: [private key block]
reason: "The guard's own test uses a PEM header string as a sample."
- repo: windy-code
hashes: ["23f32607"]
reason: "VS Code OSS extensions' package.json aiKey: Microsoft's public telemetry (App Insights) key, shipped in every VS Code build; not a Windy credential."
- repo: windytalk
hashes: ["c4189d79"]
reason: "apps/desktop/test/diagnostics.test.ts redaction fixture (hexSecret beside a fake sk-ant token); hash checked against the lockbox 10-01: not present."

View File

@@ -7,6 +7,7 @@
<div class="wg-cta">
{{if not .IsSigned}}
<a class="ui primary button" href="{{AppSubUrl}}/user/login">Sign in with Windy</a>
<p class="wg-note" style="margin-top:.9rem;font-size:.95rem;opacity:.75">Invite only at launch: new accounts are not open yet.</p>
{{else}}
<a class="ui primary button" href="{{AppSubUrl}}/{{.SignedUser.Name}}?tab=repositories">Your repositories</a>
{{end}}
@@ -18,8 +19,8 @@
<p>Git and LFS over Windy Cloud storage. Source, weights and adapters live side by side.</p></div>
<div class="wg-card"><h3>Agents are first-class</h3>
<p>An agent signs in with its own Eternitas passport — not a human's borrowed token — and its work is attributable to it.</p></div>
<div class="wg-card"><h3>Your account, no second password</h3>
<p>Sign in with the Windy account you already have. Nothing new to remember.</p></div>
<div class="wg-card"><h3>Invite only, for now</h3>
<p>Windy Git accounts are by invitation while we launch. If you've been invited, you sign in with your Windy account, with no second password.</p></div>
<div class="wg-card"><h3>Kept, and kept elsewhere</h3>
<p>Every repository is bundled nightly to off-site storage, and the restore is rehearsed rather than assumed.</p></div>
</div>

View File

@@ -0,0 +1,61 @@
{{/* Windy Git override of Gitea 1.24.6 templates/user/auth/link_account.tmpl.
Forge registration is CLOSED at launch (Grant, 09-23), so a Windy account
with no forge account used to land on Gitea's raw "Registration is disabled"
error. That case now gets a plain invite-only page; every other case is
Gitea's template unchanged (re-diff it on Gitea upgrades). No change to who
can register. */}}
{{if and .DisableRegistration (not .user_exists)}}
{{template "base/head" .}}
<div role="main" aria-label="Windy Git is invite-only" class="page-content user link-account">
<div class="ui middle very relaxed page grid">
<div class="column tw-my-5">
<div class="tw-flex tw-flex-col tw-gap-4 tw-max-w-2xl tw-m-auto">
<h4 class="ui top attached header center">Windy Git is invite-only while we launch</h4>
<div class="ui attached segment">
<p>You're signed in with your Windy account{{if or .user_name .email}} as <strong>{{or .user_name .email}}</strong>{{end}}. Nothing is wrong with it: Windy Git just isn't open to every account yet.</p>
<p>We'll let you know when it opens.</p>
<a class="ui primary button" href="https://app.windyword.ai/dashboard">Back to your dashboard</a>
</div>
</div>
</div>
</div>
</div>
{{template "base/footer" .}}
{{else}}
{{template "base/head" .}}
<div role="main" aria-label="{{.Title}}" class="page-content user link-account">
<overflow-menu class="ui secondary pointing tabular top attached borderless menu secondary-nav">
<div class="overflow-menu-items tw-justify-center">
<!-- TODO handle .ShowRegistrationButton once other login bugs are fixed -->
{{if not .AllowOnlyInternalRegistration}}
<a class="item {{if not .user_exists}}active{{end}}"
data-tab="auth-link-signup-tab">
{{ctx.Locale.Tr "auth.oauth_signup_tab"}}
</a>
{{end}}
<a class="item {{if .user_exists}}active{{end}}"
data-tab="auth-link-signin-tab">
{{ctx.Locale.Tr "auth.oauth_signin_tab"}}
</a>
</div>
</overflow-menu>
<div class="ui middle very relaxed page grid">
<div class="column tw-my-5">
{{/* these styles are quite tricky but it needs to be the same as the signin page */}}
<div class="ui tab {{if not .user_exists}}active{{end}}" data-tab="auth-link-signup-tab">
<div class="tw-flex tw-flex-col tw-gap-4 tw-max-w-2xl tw-m-auto">
{{if .AutoRegistrationFailedPrompt}}<div class="ui message">{{.AutoRegistrationFailedPrompt}}</div>{{end}}
{{template "user/auth/signup_inner" .}}
</div>
</div>
<div class="ui tab {{if .user_exists}}active{{end}}" data-tab="auth-link-signin-tab">
<div class="tw-flex tw-flex-col tw-gap-4 tw-max-w-2xl tw-m-auto">
{{template "user/auth/signin_inner" .}}
</div>
</div>
</div>
</div>
</div>
{{template "base/footer" .}}
{{end}}

View File

@@ -0,0 +1,3 @@
[Service]
# Orchestrator 09-23: compute-guard BLOCKS lane-owned findings; Grant-owned (ci/grant-owned.yml) stay WARN.
Environment=COMPUTE_GUARD_MODE=block

View File

@@ -0,0 +1,3 @@
[Service]
# Orchestrator 09-24: secret-guard BLOCKS lane-owned findings; Grant-owned stay WARN.
Environment=SECRET_GUARD_MODE=block

View File

@@ -110,7 +110,12 @@ their CI permanently, not a stopgap:
- **Image-build jobs** (name matches `docker`) post nothing: job containers
have no Docker daemon by design (I-5), so they are red on every commit. A
rootless builder (BuildKit rootless / buildx in the capped dind) is the open
decision that would bring them back.
decision that would bring them back. Jobs that need Docker but are named otherwise go in
`BRIDGE_NO_DAEMON` (empty since eternitas converted to ci/smoke, #179). DECIDED 09-23 (orchestrator,
option A): lanes convert these jobs to no-Docker smoke tests (job `services:` +
start the app + curl /health); the real image build is the deploy step on the
target host. ci-hygiene flags docker build/compose/run in workflows ("needs docker").
No host Docker socket for CI without a separate decision.
**Onboarding another private repo** — the promotion steps below, then:

258
scripts/ci_hygiene.py Normal file
View File

@@ -0,0 +1,258 @@
#!/usr/bin/env python3
"""CI hygiene guard: installs come from a lockfile, never "latest" (house rule 6).
A floating install lets CI test different versions than prod ships, and a
rebuild silently changes prod. Windy Cloud's OpenAPI test failed on exactly
that (fastapi 0.141.1 in CI vs 0.136.0 on the dev box) and all three Cloud
cells floated in prod. Also flags services that publish a HOST port: every
CI job shares one dind daemon, so two jobs publishing 5432 collide ("port is
already allocated", Windy Mind runs 147/176).
WARN-ONLY (`windy-git/ci-hygiene`, green + "⚠ WARN"); CI_HYGIENE_MODE=block
turns it red once the lanes report clean. Scans CI workflow files and
Dockerfiles only. PR heads: lines the PR adds. Default branch: every line.
OK (not flagged):
pip / uv pip install -r FILE (with or without --require-hashes), --no-deps,
exact pins (tool==1.2.3), pip/setuptools/wheel upgrades
uv sync --locked | --frozen npm ci
npm install pkg@1.2.3 (every package exact-pinned)
yarn install --frozen-lockfile / --immutable pnpm install --frozen-lockfile
Also flagged: `:latest` images (FROM / COPY --from / image: / docker://) and
`COPY uv.lock* ...`-style globs that build without the lock (Windy Mail #147).
Exceptions: ci/ci-hygiene-allow.yml, one reason per entry.
python3 scripts/ci_hygiene.py report [repo ...]
"""
from __future__ import annotations
import hashlib
import json
import os
import re
import shlex
import subprocess
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).resolve().parent))
import compute_guard as cg # noqa: E402 (shared walker, cache and allow-list loader)
ROOT = Path(__file__).resolve().parents[1]
ALLOW_FILE = Path(os.environ.get("CI_HYGIENE_ALLOW", ROOT / "ci" / "ci-hygiene-allow.yml"))
MODE = os.environ.get("CI_HYGIENE_MODE", "warn")
# CI workflow files and Dockerfiles; never vendored copies.
INCLUDE = re.compile(r"(^|/)\.(github|gitea)/workflows/[^/]+\.ya?ml$|(^|/)(Dockerfile[^/]*|[^/]+\.Dockerfile)$")
NEVER = re.compile(r"(^|/)(node_modules|vendor|third_party)/")
PREFILTER = (r"pip3? install|pip install|uv sync|npm (install|i )|yarn install|pnpm install"
r"|^\s*-\s*['\"]?[0-9]+:[0-9]+|:latest|lock[^ ]*\*"
r"|docker[ -]compose|docker (build|buildx|run)|docker/build-push-action")
TOOLING = {"pip", "setuptools", "wheel"}
NO_DOCKER_FIX = "use job services: + a no-Docker smoke test; the image builds at deploy"
DOCKER_FILE = re.compile(r"(^|/)(Dockerfile[^/]*|[^/]+\.Dockerfile)$")
LATEST = re.compile(r"(?:^\s*FROM\s+(?:--platform=\S+\s+)?|--from=|image:\s*['\"]?|docker://)([\w./-]+):latest\b", re.I)
LOCKNAME = re.compile(r"(uv\.lock|poetry\.lock|package-lock\.json|pnpm-lock\.yaml|yarn\.lock|requirements[^ ]*\.(txt|lock))", re.I)
EXACT_PY = re.compile(r"^[A-Za-z0-9._-]+(\[[^\]]*\])?==[A-Za-z0-9.+!-]+$")
EXACT_NPM = re.compile(r"^(@[^/@]+/)?[^/@]+@\d+\.\d+\.\d+([-+][0-9A-Za-z.-]+)?$")
HOST_PORT = re.compile(r"^\s*-\s*['\"]?(\d{2,5}):(\d{2,5})['\"]?\s*(#.*)?$")
PIP_VALUE_FLAGS = {"-c", "--constraint", "-i", "--index-url", "--extra-index-url", "-f",
"--find-links", "--target", "-t", "--python", "--prefix", "--root", "--platform",
"--python-version", "--implementation", "--abi", "--only-binary", "--no-binary"}
def path_ok(path: str) -> bool:
return bool(INCLUDE.search(path)) and not NEVER.search(path)
def _commands(text: str) -> list[list[str]]:
"""Split a shell line into simple commands (&&, ||, ;, |), tokenized."""
out = []
for part in re.split(r"&&|\|\||;|\|", text):
try:
toks = shlex.split(part, comments=True)
except ValueError:
toks = part.split()
# Dockerfile RUN prefix / sudo / env-prefixed assignments
while toks and (toks[0] in ("RUN", "sudo", "exec", "-", "run:", "command:")
or re.match(r"^[A-Z_][A-Z0-9_]*=", toks[0])):
toks = toks[1:]
if toks:
out.append(toks)
return out
def _pip_problem(args: list[str]) -> str | None:
if "-r" in args or "--requirement" in args or any(a.startswith("--requirement=") for a in args):
return None
if "--no-deps" in args:
return None
pkgs, skip = [], False
for a in args:
if skip:
skip = False
continue
if a in PIP_VALUE_FLAGS:
skip = True
continue
if a.startswith("-") and a not in ("-e", "--editable"):
continue
if a in ("-e", "--editable"):
continue
pkgs.append(a)
loose = [p for p in pkgs if not EXACT_PY.match(p) and p.split("[")[0].lower() not in TOOLING]
if loose:
return f"floating pip install: {' '.join(loose)[:40]}"
return None
def scan_line(path: str, text: str) -> list[tuple[str, str]]:
if cg.COMMENT.match(text):
return []
hits = []
if "/workflows/" in path and HOST_PORT.match(text):
hits.append(("host port", f"service publishes host port {HOST_PORT.match(text).group(1)} (shared dind)"))
return hits
# Windy Mail #147: a `:latest` build/tool image floats exactly like an
# unpinned package, and `COPY uv.lock* ./` builds WITHOUT the lock when it
# is missing instead of failing.
m = LATEST.search(text)
if m:
hits.append(("floating image", f"{m.group(1)}:latest"))
if DOCKER_FILE.search(path) and re.match(r"^\s*COPY\b", text, re.I):
globbed = [t for t in text.split() if "*" in t and LOCKNAME.search(t)]
if globbed:
hits.append(("optional lock", f"COPY {globbed[0]} (must fail if the lock is missing)"))
# Windy Git jobs get NO Docker daemon (I-5), so a docker build/compose/run
# step in CI can never pass here (orchestrator 09-23, option A). The real
# image build is the deploy step on the target host.
if "/workflows/" in path and re.search(r"uses:\s*['\"]?docker/build-push-action", text):
hits.append(("needs docker", "docker/build-push-action in CI (no Docker daemon on Windy Git; " + NO_DOCKER_FIX + ")"))
for toks in _commands(text):
low = [t.lower() for t in toks]
if "/workflows/" in path and (
low[:2] in (["docker", "build"], ["docker", "buildx"], ["docker", "run"], ["docker", "compose"])
or low[:1] == ["docker-compose"]
):
hits.append(("needs docker", f"{' '.join(low[:2])} in CI (no Docker daemon on Windy Git; " + NO_DOCKER_FIX + ")"))
continue
# pip install / python -m pip install / uv pip install
for i in range(len(low) - 1):
if os.path.basename(low[i]) in ("pip", "pip3") and low[i + 1] == "install":
prob = _pip_problem(toks[i + 2:])
if prob:
hits.append(("floating install", prob))
break
if low[:2] == ["uv", "sync"] and not ({"--locked", "--frozen"} & set(low)):
hits.append(("floating install", "uv sync without --locked/--frozen"))
if low[:1] == ["npm"] and len(low) > 1 and low[1] in ("install", "i", "add"):
pkgs = [t for t in toks[2:] if not t.startswith("-")]
if not pkgs or not all(EXACT_NPM.match(p) for p in pkgs):
hits.append(("floating install", f"npm {low[1]} {' '.join(pkgs)[:30]}".strip() + " (use npm ci)"))
if low[:2] == ["yarn", "install"] and not ({"--frozen-lockfile", "--immutable"} & set(low)):
hits.append(("floating install", "yarn install without --frozen-lockfile"))
if low[:2] == ["pnpm", "install"] and "--frozen-lockfile" not in low:
hits.append(("floating install", "pnpm install without --frozen-lockfile"))
return hits
_DISABLED: dict[str, set[str]] | None = None
def disabled_workflows() -> dict[str, set[str]]:
"""Workflow file names Gitea has DISABLED per repo (lowercased repo name).
Deploy/release workflows are disabled on Windy Git: they run on the target
host, where a Docker daemon really exists, so "needs docker" must not flag
them. One bounded query per process; on any failure nothing is excused
(flag rather than hide).
"""
global _DISABLED
if _DISABLED is not None:
return _DISABLED
_DISABLED = {}
query = ("select coalesce(json_object_agg(r.lower_name, u.config::json->'DisabledWorkflows'), '{}'::json)"
" from repo_unit u join repository r on r.id = u.repo_id"
" where u.type = 10 and u.config like '%DisabledWorkflows%';")
try:
out = subprocess.run(
["docker", "exec", "-i", "windy-git-db-1", "sh", "-c",
'psql -U "$POSTGRES_USER" -d gitea -At -v ON_ERROR_STOP=1'],
input=query, capture_output=True, text=True, check=True, timeout=30,
).stdout.strip()
_DISABLED = {k: set(v or []) for k, v in json.loads(out or "{}").items()}
except (subprocess.SubprocessError, OSError, ValueError):
pass
return _DISABLED
def _runs_here(repo: str, findings):
"""Drop "needs docker" hits in workflows that never run on Windy Git."""
if findings is None:
return None
off = disabled_workflows().get(repo.lower(), set())
return [f for f in findings if not (f.kind == "needs docker" and Path(f.path).name in off)]
def check(repo: str, sha: str, default_branch: str, is_default_head: bool):
return _runs_here(repo, _check(repo, sha, default_branch, is_default_head))
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 = cg.load_allow(ALLOW_FILE)
rules = hashlib.sha256((PREFILTER + INCLUDE.pattern + EXACT_PY.pattern + EXACT_NPM.pattern).encode()).hexdigest()[:8]
fp = cg._fingerprint(allow) + ":" + rules # hashlib, not hash(): hash() is per-process random
kw = dict(line_fn=scan_line, path_ok=path_ok)
if is_default_head:
return cg.cached_scan(f"hyg-tree:{repo}:{sha}:{fp}",
lambda: cg.scan_tree(repo, bare, sha, allow, prefilter=PREFILTER, **kw))
return cg.cached_scan(f"hyg-pr:{repo}:{sha}:{fp}",
lambda: cg.scan_added(repo, bare, f"refs/heads/{default_branch}", sha, allow, **kw))
def status_for(findings, whole_tree: bool, grant=()):
"""Same contract as compute_guard.status_for: `grant` findings never block."""
scope = "in CI/Dockerfiles" if whole_tree else "added"
if not findings and grant:
g, n = grant[0], len(grant)
desc = f"⚠ WARN (Grant-owned, not blocking): {n} CI hygiene issue{'s' if n > 1 else ''} {scope}, e.g. {g.path}:{g.line} {g.match}"
return "success", desc[:140], g
if not findings:
return "success", f"OK: no floating install or host-port service {scope}", None
f = findings[0]
n = len(findings)
state = "failure" if MODE == "block" else "success"
lead = "BLOCKED" if MODE == "block" else "⚠ WARN (not blocking)"
return state, f"{lead}: {n} CI hygiene issue{'s' if n > 1 else ''} {scope}, e.g. {f.path}:{f.line} {f.match}"[:140], f
def report(repos: list[str]) -> int:
allow = cg.load_allow(ALLOW_FILE)
total = 0
for repo in repos:
bare = cg.WORK / f"{repo}.git"
if not bare.is_dir():
print(f"## {repo}: no sync clone, skipped")
continue
head = cg._git(bare, "symbolic-ref", "--short", "HEAD").strip()
sha = cg._git(bare, "rev-parse", head).strip()
fs = _runs_here(repo, cg.scan_tree(repo, bare, sha, allow, line_fn=scan_line, path_ok=path_ok, prefilter=PREFILTER))
total += len(fs)
print(f"## {repo} ({head} {sha[:7]}): {len(fs)} issue(s)")
for f in fs:
print(f" {f.path}:{f.line} [{f.kind}] {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__)

View File

@@ -98,9 +98,18 @@ def load_allow(path: Path = ALLOW_FILE) -> list[dict]:
return entries
def allowed(repo: str, path: str, allow: list[dict]) -> bool:
def allowed(repo: str, path: str, allow: list[dict], text: str | None = None) -> bool:
"""An entry may carry `matches:` (regexes): then only lines matching one of
them are allowed, so an allowed file can't smuggle in a NEW call (e.g. an
OAuth sign-in endpoint is allowed, an inference endpoint in the same file
still flags). Entries without `matches` cover the whole path."""
for e in allow:
if e["repo"] == repo and any(fnmatch.fnmatch(path, g) for g in e["paths"]):
if e["repo"] != repo or not any(fnmatch.fnmatch(path, g) for g in e["paths"]):
continue
pats = e.get("matches")
if not pats:
return True
if text is not None and any(re.search(rx, text) for rx in pats):
return True
return False
@@ -130,11 +139,20 @@ def _git(bare: Path, *args: str) -> str:
).stdout
def scan_tree(repo: str, bare: Path, sha: str, allow: list[dict]) -> list[Finding]:
def _default_path_ok(path: str) -> bool:
return not SKIP.search(path)
def scan_tree(repo: str, bare: Path, sha: str, allow: list[dict], *, line_fn=None,
path_ok=None, prefilter: str | None = None) -> list[Finding]:
"""Every line in the tree at `sha` (default branch: the baseline)."""
# A cheap prefilter by git, then the real rules in Python.
pre = "|".join([re.escape(h) for h in HOSTS] + KEYS + ["anthropic", "openai", "groq", "mistral",
"generativeai", "genai", "cohere", "together", "cerebras", "litellm"])
# Other guards (ci_hygiene) reuse this walker with their own line rules.
line_fn = line_fn or scan_line
path_ok = path_ok or _default_path_ok
pre = prefilter or "|".join([re.escape(h) for h in HOSTS] + KEYS + [
"anthropic", "openai", "groq", "mistral", "generativeai", "genai", "cohere",
"together", "cerebras", "litellm"])
try:
out = _git(bare, "grep", "-nIE", "-e", pre, sha, "--", ".")
except subprocess.CalledProcessError as e:
@@ -148,24 +166,26 @@ def scan_tree(repo: str, bare: Path, sha: str, allow: list[dict]) -> list[Findin
_, path, line, text = raw.split(":", 3)
except ValueError:
continue
if SKIP.search(path) or allowed(repo, path, allow):
if not path_ok(path) or allowed(repo, path, allow, text):
continue
for kind, match in scan_line(path, text):
for kind, match in line_fn(path, text):
found.append(Finding(path, int(line), kind, match))
return found
def scan_added(repo: str, bare: Path, base_ref: str, sha: str, allow: list[dict]) -> list[Finding]:
def scan_added(repo: str, bare: Path, base_ref: str, sha: str, allow: list[dict], **kw) -> list[Finding]:
"""Only the lines a PR adds, vs its merge-base with the default branch."""
mb = _git(bare, "merge-base", base_ref, sha).strip()
diff = _git(bare, "diff", "-U0", "--no-color", "--no-ext-diff", mb, sha)
return parse_added(repo, diff, allow)
return parse_added(repo, diff, allow, **kw)
HUNK = re.compile(r"^@@ -\d+(?:,\d+)? \+(\d+)(?:,\d+)? @@")
def parse_added(repo: str, diff: str, allow: list[dict]) -> list[Finding]:
def parse_added(repo: str, diff: str, allow: list[dict], *, line_fn=None, path_ok=None) -> list[Finding]:
line_fn = line_fn or scan_line
path_ok = path_ok or _default_path_ok
found, path, line = [], None, 0
for raw in diff.splitlines():
if raw.startswith("+++ "):
@@ -179,8 +199,8 @@ def parse_added(repo: str, diff: str, allow: list[dict]) -> list[Finding]:
if path is None or raw.startswith("--- "):
continue
if raw.startswith("+"):
if not (SKIP.search(path) or allowed(repo, path, allow)):
for kind, match in scan_line(path, raw[1:]):
if path_ok(path) and not allowed(repo, path, allow, raw[1:]):
for kind, match in line_fn(path, raw[1:]):
found.append(Finding(path, line, kind, match))
line += 1
return found
@@ -214,10 +234,21 @@ def cached_scan(key: str, fn) -> list[Finding]:
return result
def fetched(bare: Path, sha: str) -> bool:
"""Is `sha` in the sync clone yet? The bridge learns PR / default heads from
GitHub's API AFTER the sync fetched, so a push in between is simply not here
until the next 5-min cycle. That is a race, not an error: skip quietly."""
try:
_git(bare, "cat-file", "-e", f"{sha}^{{commit}}")
return True
except subprocess.CalledProcessError:
return False
def check(repo: str, sha: str, default_branch: str, is_default_head: bool) -> list[Finding] | None:
"""Findings for one commit, or None when the guard can't run (never a fake OK)."""
bare = WORK / f"{repo}.git"
if not bare.is_dir():
if not bare.is_dir() or not fetched(bare, sha):
return None
allow = load_allow()
fp = _fingerprint(allow)
@@ -229,9 +260,19 @@ def check(repo: str, sha: str, default_branch: str, is_default_head: bool) -> li
)
def status_for(findings: list[Finding], whole_tree: bool) -> tuple[str, str, Finding | None]:
"""(state, description, first finding) for the GitHub commit status."""
def status_for(findings: list[Finding], whole_tree: bool,
grant: list[Finding] = ()) -> tuple[str, str, Finding | None]:
"""(state, description, first finding) for the GitHub commit status.
`findings` = lane-owned (these block in MODE=block); `grant` = findings in
Grant-owned code (ci/grant-owned.yml): always WARN, never red (orchestrator
09-23: his desktop work is never blocked by us)."""
scope = "in tree" if whole_tree else "added"
if not findings and grant:
g, n = grant[0], len(grant)
desc = (f"⚠ WARN (Grant-owned, not blocking): {n} direct AI-provider use{'s' if n > 1 else ''} "
f"{scope}, e.g. {g.path}:{g.line} {g.match}")
return "success", desc[:140], g
if not findings:
what = "no direct AI-provider use in tree" if whole_tree else "no direct AI-provider use added"
return "success", f"OK: {what} (Windy Mind is the only door)", None

161
scripts/guards_report.py Normal file
View File

@@ -0,0 +1,161 @@
#!/usr/bin/env python3
"""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:
windy-pro's desktop app and its build jobs) are listed in their OWN section and
do not count against "ready to block": those are proposals for Grant, not a
lane's fix (orchestrator, 09-23).
sudo python3 scripts/guards_report.py > GUARDS_STATUS.md
"""
from __future__ import annotations
import fnmatch
import os
import re
import sys
import time
from pathlib import Path
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"))
REPOS = os.environ.get("BRIDGE_REPOS", "").split() or [
"windy-chat", "windy-mail", "windy-calendar", "Windy-Clone", "WindyCloud", "windy-search",
"windy-connect", "windy-drops", "windy-code-web", "windy-code", "windy-traveler",
"windy-registry", "eternitas", "windy-translate", "windytranslate-site", "windytraveler-site",
"windy-hand", "windy-cloud-sites", "windy-cloud-domains", "windy-cloud-vps", "windytalk",
"windy-pro", "windy-mind", "windy-git", "windy-inbox"]
# Owner lane per repo = the session name to message (orchestrator routing, 09-23 ~22:45Z:
# hub/account-server/dashboard/site -> "Windy Hub"; windy-calendar only -> "Windy Calender";
# windy-admin/telemetry -> "Windy Admin"). windy-pro here = its server/web side; the desktop
# app is Grant-owned and listed in its own section below.
OWNERS = {
"windy-chat": "Windy Chat", "windy-mail": "Windy Mail", "windy-calendar": "Windy Calender",
"Windy-Clone": "Windy Clone", "WindyCloud": "Windy Cloud", "windy-cloud-sites": "Windy Cloud",
"windy-cloud-domains": "Windy Cloud", "windy-cloud-vps": "Windy Cloud", "windy-search": "Windy Search",
"windy-connect": "Windy Connect", "windy-drops": "Windy Drops", "windy-registry": "Windy Drops",
"windy-code-web": "Windy Code", "windy-code": "Windy Code", "windy-traveler": "Windy Traveler",
"windytraveler-site": "Windy Traveler", "eternitas": "Eternitas", "windy-translate": "Windy Translate",
"windytranslate-site": "Windy Translate", "windy-hand": "Windy Hand", "windytalk": "Windy Talk",
"windy-pro": "Windy Hub", "windy-mind": "WIndy Mind", "windy-git": "Windy Git",
"windy-inbox": "Windy Drops"} # Windy Inbox build lead (09-24)
JOB = re.compile(r"^ ([A-Za-z0-9_-]+):\s*$")
def job_of(text: str, line: int) -> str | None:
"""The workflow job a line belongs to (2-space keys under `jobs:`)."""
in_jobs, job = False, None
for i, raw in enumerate(text.splitlines(), 1):
if raw.startswith("jobs:"):
in_jobs = True
elif in_jobs and JOB.match(raw):
job = JOB.match(raw).group(1)
elif raw and not raw[0].isspace() and not raw.startswith("jobs:"):
in_jobs = False
if i == line:
return job if in_jobs else None
return None
def grant_owned(repo: str, path: str, job: str | None, owned: list[dict]) -> bool:
for e in owned:
if e["repo"] != repo:
continue
if any(fnmatch.fnmatch(path, g) for g in e.get("paths") or []):
return True
if job and job in (e.get("jobs") or {}).get(path, []):
return True
return False
def split_grant(repo: str, sha: str, findings: list) -> tuple[list, list]:
"""(lane-owned, Grant-owned) findings at `sha`, by ci/grant-owned.yml, with
workflow lines attributed to their job exactly as the status page does."""
owned = (yaml.safe_load(OWNED.read_text()) or {}).get("grant_owned") or []
if not any(e["repo"] == repo for e in owned):
return list(findings), []
bare = cg.WORK / f"{repo}.git"
texts: dict[str, str] = {}
lane, grant = [], []
for f in findings:
job = None
if "/workflows/" in f.path:
if f.path not in texts:
texts[f.path] = cg._git(bare, "show", f"{sha}:{f.path}")
job = job_of(texts[f.path], f.line)
(grant if grant_owned(repo, f.path, job, owned) else lane).append(f)
return lane, grant
def scan(repo: str, owned: list[dict]):
bare = cg.WORK / f"{repo}.git"
if not bare.is_dir():
return None
head = cg._git(bare, "symbolic-ref", "--short", "HEAD").strip()
sha = cg._git(bare, "rev-parse", head).strip()
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 []),
("secrets", sgd.check(repo, sha, head, True) or [])):
for f in fs:
job = None
if "/workflows/" in f.path:
if f.path not in texts:
texts[f.path] = cg._git(bare, "show", f"{sha}:{f.path}")
job = job_of(texts[f.path], f.line)
out[key].append((f, job, grant_owned(repo, f.path, job, owned)))
return out
def render(results: dict) -> str:
now = time.strftime("%Y-%m-%d %H:%MZ", time.gmtime())
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.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'} |",
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]]
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", "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"
def main() -> int:
owned = (yaml.safe_load(OWNED.read_text()) or {}).get("grant_owned") or []
results = {}
for repo in REPOS:
r = scan(repo, owned)
if r is not None:
results[repo] = r
sys.stdout.write(render(results))
return 0
if __name__ == "__main__":
sys.exit(main())

View File

@@ -53,7 +53,8 @@ REPOS = os.environ.get(
"windy-chat windy-mail windy-calendar Windy-Clone WindyCloud windy-search windy-connect"
" windy-drops windy-code-web windy-code windy-traveler windy-registry eternitas"
" windy-translate windytranslate-site windytraveler-site windy-hand"
" windy-cloud-sites windy-cloud-domains windy-cloud-vps windytalk windy-pro windy-mind",
" windy-cloud-sites windy-cloud-domains windy-cloud-vps windytalk windy-pro windy-mind"
" windy-inbox windy-text windy-call windy-cell",
).split()
# Gitea run status -> GitHub status state. `skipped` is deliberately absent: a
@@ -74,7 +75,23 @@ MIRROR_TAG = "[GH#"
# Posting them would put a permanent red X on every commit, and a signal that is
# always red trains everyone to ignore red. Not posted until a rootless builder
# exists; that is a decision, recorded in docs/CUTOVER.md, not a failure.
NO_DAEMON_JOB = re.compile(r"docker", re.IGNORECASE)
# ...but NOT the no-Docker smoke jobs that replaced them (option A, 09-23):
# windy-search's "Boot smoke (no Docker)" matched plain `docker` and was hidden.
NO_DAEMON_JOB = re.compile(r"(?<!no )(?<!no-)(?<!without )docker", re.IGNORECASE)
# Image-build jobs whose NAME doesn't say docker (orchestrator 09-23, option A:
# each lane converts the job to a no-Docker smoke test; until then it is not
# posted). Format: "repo:workflow/job,...;repo2:...".
NO_DAEMON_NAMED: dict[str, set[str]] = {}
# eternitas:ci/build dropped 09-23: converted to a no-Docker ci/smoke (#179).
for _entry in os.environ.get("BRIDGE_NO_DAEMON", "").split(";"):
if ":" in _entry:
_repo, _jobs = _entry.split(":", 1)
NO_DAEMON_NAMED[_repo.strip()] = {j.strip() for j in _jobs.split(",") if j.strip()}
def needs_daemon(repo: str, wf: str, job: str) -> bool:
"""True for image-build jobs, which cannot run here (no Docker daemon, I-5)."""
return bool(NO_DAEMON_JOB.search(job)) or f"{wf}/{job}" in NO_DAEMON_NAMED.get(repo, ())
# Jobs Grant ruled NON-BLOCKING (GRANT_DECISIONS_2026-09-23): still run on
# Windy Git and visible there, but not posted to GitHub, so they cannot turn a
@@ -202,7 +219,17 @@ def sync_prs(repo: str) -> list[str]:
wanted.add(tag)
heads.append(pr["head"]["sha"])
if tag in ours:
cur = (ours[tag].get("base") or {}).get("ref")
if cur is None or cur == pr["base"]["ref"]: # unknown base: never guess, leave it
continue
# Retargeted on GitHub (e.g. a stacked PR moved to main after its
# parent merged). The mirror kept the OLD base, so workflows filtered
# on the base (`pull_request: branches: [main]`) silently stopped
# running: eternitas #167, 09-23. Replace the mirror: an "edited"
# event triggers nothing, a freshly opened PR runs CI at once.
gitea("PATCH", f"/repos/{WG_OWNER}/{repo}/pulls/{ours[tag]['number']}", {"state": "closed"})
print(f" {repo}: GH#{pr['number']} retargeted {cur} -> "
f"{pr['base']['ref']}: replacing its mirror PR")
st, _ = gitea(
"POST",
f"/repos/{WG_OWNER}/{repo}/pulls",
@@ -284,7 +311,7 @@ def post_statuses(repo: str, sha: str) -> None:
break
latest: dict[str, dict] = {}
for r in runs:
if r["head_sha"] != sha or NO_DAEMON_JOB.search(r["name"]):
if r["head_sha"] != sha or needs_daemon(repo, r["workflow_id"].removesuffix(".yml"), r["name"]):
continue
if f"{r['workflow_id'].removesuffix('.yml')}/{r['name']}" in NON_BLOCKING.get(repo, ()):
continue
@@ -294,9 +321,9 @@ def post_statuses(repo: str, sha: str) -> None:
# Queued jobs: `pending` where nothing newer has been picked up. A re-run
# queued behind an old failure must read pending, not the stale red.
for q in queued_jobs(repo, sha):
if NO_DAEMON_JOB.search(q["name"]):
continue
wf = q["workflow_id"].removesuffix(".yml")
if needs_daemon(repo, wf, q["name"]):
continue
if f"{wf}/{q['name']}" in NON_BLOCKING.get(repo, ()):
continue
ctx = f"windy-git/{wf}/{q['name']}"
@@ -344,34 +371,58 @@ 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:
"""Windy Mind is the only door to AI compute: flag direct provider use (warn-only).
"""Windy Mind is the only door to AI compute: flag direct provider use (warn-only)."""
_post_guard("compute_guard", GUARD_CTX, repo, sha, default_branch, is_default_head)
Non-fatal and never a fake OK: if the guard can't run, nothing is posted.
"""
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)
def _post_guard(modname: str, ctx: str, repo: str, sha: str, default_branch: str,
is_default_head: bool) -> None:
"""One code path for every repo-scanning guard. Non-fatal and never a fake OK:
if the guard can't run, nothing is posted."""
try:
import compute_guard as cg # same directory; loaded lazily so the bridge never depends on it
import importlib
findings = cg.check(repo, sha, default_branch, is_default_head)
except Exception as e: # noqa: BLE001 — the guard must never break CI signals
print(f" {repo}@{sha[:7]} compute-guard skipped ({type(e).__name__}: {str(e)[:80]})")
g = importlib.import_module(modname) # same directory; lazy so the bridge never depends on it
findings = g.check(repo, sha, default_branch, is_default_head)
except Exception as e: # noqa: BLE001 — a guard must never break CI signals
print(f" {repo}@{sha[:7]} {ctx} skipped ({type(e).__name__}: {str(e)[:80]})")
return
if findings is None:
return
state, desc, first = cg.status_for(findings, whole_tree=is_default_head)
try: # Grant-owned code never blocks (orchestrator 09-23); lazy like the guards
import importlib
lane, grant = importlib.import_module("guards_report").split_grant(repo, sha, findings)
except Exception as e: # noqa: BLE001 — can't tell whose code: warn, never block
print(f" {repo}@{sha[:7]} {ctx}: Grant-owned split failed ({type(e).__name__}); WARN only")
lane, grant = [], list(findings)
state, desc, first = g.status_for(lane, whole_tree=is_default_head, grant=grant)
st, existing = github("GET", f"/repos/{GH_OWNER}/{repo}/commits/{sha}/statuses?per_page=100")
for s in existing or []: # newest first: compare the latest guard status only
if s["context"] == GUARD_CTX:
if s["context"] == ctx:
if (s["state"], s.get("description")) == (state, desc):
return
break
url = (f"{PUBLIC}/{WG_OWNER}/{repo}/src/commit/{sha}/{first.path}#L{first.line}"
if first else f"{PUBLIC}/{WG_OWNER}/{repo}/src/commit/{sha}")
st, _ = github("POST", f"/repos/{GH_OWNER}/{repo}/statuses/{sha}",
{"state": state, "context": GUARD_CTX, "description": desc, "target_url": url})
print(f" {repo}@{sha[:7]} {GUARD_CTX} = {state} ({len(findings)} finding(s)) -> {st}")
{"state": state, "context": ctx, "description": desc, "target_url": url})
print(f" {repo}@{sha[:7]} {ctx} = {state} ({len(findings)} finding(s)) -> {st}")
def main() -> int:
@@ -392,6 +443,8 @@ def main() -> int:
for sha in dict.fromkeys(shas):
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

View File

@@ -0,0 +1,118 @@
#!/usr/bin/env python3
"""Weekly: every PUBLIC repo in Grant's GitHub accounts, full history (every object,
reachable or not, incl. PR refs), for secret-shaped strings. Leak hunt 09-24: a
real bot token sat in a public test fixture for five months.
Output is hash + location only, never a value (house rule 10). Known fakes are
excused by ci/secret-guard-allow.yml (same file as secret-guard; the repo NAME is
matched across accounts). Runs on Veron as user1-gpu (gh is logged in there):
python3 scripts/public_secret_scan.py [--out FILE] [--owners a,b,...]
"""
from __future__ import annotations
import argparse
import fnmatch
import json
import subprocess
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).resolve().parent))
import secret_shapes as ss # noqa: E402
OWNERS = ["sneakyfree", "VERONTECH", "Windstorm-Institute", "Windstorm-Labs", "Public-Streamer"]
WORK = Path.home() / "leakscan" / "public"
MAX_BLOB = 20_000_000
def _run(*a: str) -> subprocess.CompletedProcess:
return subprocess.run(a, capture_output=True)
def blob_hits(bare: Path) -> dict[str, list[tuple[str, str]]]:
"""{blob: [(kind, hash8)]} over every blob object in the repo."""
chk = _run("git", "-C", str(bare), "cat-file", "--batch-all-objects",
"--batch-check=%(objectname) %(objecttype) %(objectsize)")
blobs = [p[0] for p in (ln.split() for ln in chk.stdout.decode().splitlines())
if len(p) == 3 and p[1] == "blob" and int(p[2]) < MAX_BLOB]
if not blobs:
return {}
import threading
p = subprocess.Popen(["git", "-C", str(bare), "cat-file", "--batch"], stdin=subprocess.PIPE, stdout=subprocess.PIPE)
def feed() -> None: # separate thread: writing everything first deadlocks (both pipes fill)
p.stdin.write(("\n".join(blobs) + "\n").encode())
p.stdin.close()
threading.Thread(target=feed, daemon=True).start()
out: dict[str, list[tuple[str, str]]] = {}
for _ in blobs:
hdr = p.stdout.readline().split()
data = p.stdout.read(int(hdr[2]))
p.stdout.read(1)
found = ss.find(data.decode("utf-8", "ignore"))
if found:
out[hdr[0].decode()] = found
p.wait()
return out
def locate(bare: Path, blob: str) -> dict:
lg = _run("git", "-C", str(bare), "log", "--all", "--format=@@%H %cI", "--name-only",
f"--find-object={blob}").stdout.decode()
commits, paths = [], set()
for line in lg.splitlines():
if line.startswith("@@"):
commits.append(line[2:].split())
elif line.strip():
paths.add(line.strip())
head = _run("git", "-C", str(bare), "ls-tree", "-r", "HEAD").stdout.decode()
first = commits[-1] if commits else ["unreachable", "-"]
return {"paths": sorted(paths), "first_commit": first[0][:10], "first_date": first[1],
"in_head": blob in head}
def excused(repo: str, kind: str, h: str, paths: list[str], allow: dict) -> bool:
a = allow.get(repo) or {"hashes": set(), "paths": []}
if kind in {"private key block"}:
return bool(paths) and all(any(kind in k and fnmatch.fnmatch(p, g) for g, k in a["paths"]) for p in paths)
return h in a["hashes"]
def main() -> int:
ap = argparse.ArgumentParser()
ap.add_argument("--out", default=str(WORK / "latest.json"))
ap.add_argument("--owners", default=",".join(OWNERS))
args = ap.parse_args()
import secret_guard as sg
allow = sg.load_allow()
WORK.mkdir(parents=True, exist_ok=True)
rows, scanned, errors = [], [], []
for owner in args.owners.split(","):
lst = _run("gh", "repo", "list", owner, "--limit", "1000", "--visibility", "public", "--json", "name")
for r in json.loads(lst.stdout or b"[]"):
name = r["name"]
bare = WORK / owner / f"{name}.git"
if bare.exists():
c = _run("git", "-C", str(bare), "remote", "update", "--prune")
else:
bare.parent.mkdir(parents=True, exist_ok=True)
c = _run("gh", "repo", "clone", f"{owner}/{name}", str(bare), "--", "--mirror", "-q")
if c.returncode:
errors.append(f"{owner}/{name}")
continue
scanned.append(f"{owner}/{name}")
for blob, found in blob_hits(bare).items():
loc = locate(bare, blob)
for kind, h in sorted(set(found)):
rows.append({"repo": f"{owner}/{name}", "kind": kind, "hash8": h, **loc,
"excused": excused(name, kind, h, loc["paths"], allow)})
Path(args.out).write_text(json.dumps({"scanned": scanned, "errors": errors, "rows": rows}, indent=1))
new = [r for r in rows if not r["excused"]]
print(f"scanned {len(scanned)} public repos, {len(errors)} errors, {len(rows)} secret-shaped, {len(new)} NOT excused")
return 0
if __name__ == "__main__":
sys.exit(main())

View File

@@ -16,7 +16,10 @@
# itself, so Windy Git is never left behind GitHub.
set -euo pipefail
repo="${1:?repo}"; branch="${2:?branch}"; want="${3:?sha prefix}"
G="sudo docker exec -u git windy-git-gitea-1 git -C /data/git/repositories/windyadmin/${repo}.git"
# wg-q lives in the INVOKING user's ~/bin; under sudo, ~ is /root.
WGQ="${WGQ:-$(getent passwd "${SUDO_USER:-$USER}" | cut -d: -f6)/bin/wg-q}"
# Gitea stores repositories LOWERCASED on disk (WindyCloud -> windycloud.git).
G="sudo docker exec -u git windy-git-gitea-1 git -C /data/git/repositories/windyadmin/${repo,,}.git"
head=$($G rev-parse "refs/heads/${branch}")
[[ "$head" == "$want"* ]] || { echo "refusing: ${branch} is at ${head:0:7}, not ${want}"; exit 1; }
@@ -42,7 +45,7 @@ until [ "$(systemctl show windygit-sync -p ExecMainStartTimestampMonotonic --val
done
echo "restored by sync: ${branch} = ${head:0:7}"
sleep 5
~/bin/wg-q <<SQL
"$WGQ" <<SQL
select ar.index, ar.workflow_id, ar.event, ar.status, to_char(to_timestamp(ar.created),'HH24:MI:SS')
from action_run ar join repository r on r.id = ar.repo_id
where r.name = '${repo}' and ar.commit_sha = '${head}' order by ar.id desc limit 6;

131
scripts/secret_guard.py Normal file
View File

@@ -0,0 +1,131 @@
#!/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 fnmatch
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")
# Kinds that only WARN (rolled out warn-first); empty = every kind blocks in block mode.
WARN_KINDS = {k for k in os.environ.get("SECRET_GUARD_WARN_KINDS", "").split(",") if k}
NEVER = re.compile(r"(^|/)(node_modules|vendor|third_party)/")
def path_ok(path: str) -> bool:
return not NEVER.search(path)
# A private-key match is only its BEGIN line, so its hash is the same everywhere:
# those are allowed by PATH (entries with `paths` + `kinds`), everything else by HASH.
PATH_ONLY_KINDS = {"private key block"}
def load_allow(path: Path = ALLOW_FILE) -> dict[str, dict]:
"""{repo: {"hashes": {hash8}, "paths": [(glob, {kind})]}}; every entry needs a reason."""
data = yaml.safe_load(path.read_text()) if path.exists() else {}
out: dict[str, dict] = {}
for e in (data or {}).get("allow") or []:
if not (e.get("repo") and (e.get("hashes") or (e.get("paths") and e.get("kinds")))
and str(e.get("reason", "")).strip()):
raise ValueError(f"allow entry needs repo, hashes (or paths + kinds) and a reason: {e}")
if e.get("paths") and not set(e["kinds"]) <= PATH_ONLY_KINDS:
raise ValueError(f"path allows are only for {sorted(PATH_ONLY_KINDS)}: {e}")
r = out.setdefault(e["repo"], {"hashes": set(), "paths": []})
r["hashes"].update(str(h) for h in e.get("hashes") or [])
r["paths"] += [(g, set(e["kinds"])) for g in e.get("paths") or []]
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, dict]):
a = allow.get(repo) or {"hashes": set(), "paths": []}
def ok(f) -> bool:
if f.kind in PATH_ONLY_KINDS:
return any(f.kind in kinds and fnmatch.fnmatch(f.path, g) for g, kinds in a["paths"])
return f.match.rsplit("#", 1)[-1] in a["hashes"]
return [f for f in findings if not ok(f)]
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)
soft = MODE != "block" or all(x.kind in WARN_KINDS for x in findings)
state = "success" if soft else "failure"
lead = "⚠ WARN (not blocking)" if soft else "BLOCKED"
if not soft:
f = next(x for x in findings if x.kind not in WARN_KINDS)
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__)

45
scripts/secret_shapes.py Normal file
View File

@@ -0,0 +1,45 @@
"""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_-])")),
# Twilio (Windy Text 10-01: a live auth token sat in test files for months). An auth token
# is a bare 32-hex with no prefix, so it is only caught when ASSIGNED to a secret-ish name.
("twilio sid/api key", re.compile(r"\b(?:AC|SK)[0-9a-f]{32}\b")),
("32-hex secret assignment", re.compile(
r"(?i)\b[a-z0-9_.-]*(?:token|secret|key|password)[a-z0-9_.-]*[\"']?\s*[:=]\s*[\"']?(?P<v>(?<![0-9a-f])[0-9a-f]{32}(?![0-9a-f]))")),
("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-----|(AC|SK)[0-9a-f]{32}|[0-9a-fA-F]{32}")
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('v') if 'v' in rx.groupindex else m.group(0))))
return out

View File

@@ -38,7 +38,7 @@ FAILED=0
# Repos Windy Git tracks FROM GitHub. Remove a repo from this list at the moment
# it flips to Windy-Git-first, or the sync will fight its authors and win.
REPOS="${SYNC_REPOS:-windy-calendar windy-search windy-registry Windy-Clone WindyCloud windy-cloud-sites windy-mind eternitas windy-agent windy-git windy-chat windy-mail windy-connect windy-drops windy-code-web windy-code windy-traveler windy-translate windytranslate-site windytraveler-site windy-hand windy-cloud-domains windy-cloud-vps windytalk windy-pro}"
REPOS="${SYNC_REPOS:-windy-calendar windy-search windy-registry Windy-Clone WindyCloud windy-cloud-sites windy-mind eternitas windy-agent windy-git windy-chat windy-mail windy-connect windy-drops windy-code-web windy-code windy-traveler windy-translate windytranslate-site windytraveler-site windy-hand windy-cloud-domains windy-cloud-vps windytalk windy-pro windy-inbox windy-text windy-call windy-cell}"
# Repos whose TAGS must not reach Windy Git. A tag push fires `on: push: tags`
# workflows; windy-pro's build-electron is a matrix over ubuntu/macos/windows-

1485
uv.lock generated Normal file

File diff suppressed because it is too large Load Diff