Files
windy-git/api/app/auth.py
Grant Whitmer d8deffe4db
All checks were successful
check / gate (push) Successful in 18s
canary / probe (push) Successful in 11s
SECURITY: close the agent-auth bypass — trust is not authentication
Verified live 2026-08-13: a forged 'alg:none' token naming a passport lifted
from the logs returned HTTP 200 as that agent. The agent path read the passport
without verifying the EPT signature, asked Eternitas 'is this passport
reputable?', and seated the caller on a yes. That answers reputation, not
possession — anyone who knows a passport number could impersonate that agent on
the public API.

The human path already failed closed for exactly this reason
(require_verified_jwt). The gate was on the wrong path: it sat AFTER the agent
branch returned. The agent path now fails closed too, BEFORE the trust lookup,
so a forged token never even reaches Eternitas. Reopens automatically when the
ES256/JWKS verifier (G3.2/G9.1) exists and this gate consults it.

Adds BEHAVIORAL tests (not string-grep): a forged alg:none token exercised
through the real get_caller must raise, not authenticate. This is the test that
would have caught the bypass; the suite had 86 source-string assertions and
zero that ran the auth decision.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-13 22:49:45 -04:00

258 lines
10 KiB
Python

"""Identity: humans, agents, and internal services (G3.2 / G3.6 / I-6).
Three caller classes, all first-class, none a bypass:
* **human** — account-server RS256 JWT (OIDC)
* **agent** — Eternitas ES256 EPT
* **service** — `X-Service-Token`, for the Cloud portal calling `/internal/*`
There is deliberately no fourth class and no escape hatch. The Windy Word desktop
control server is the best Principle-#5 artifact in the ecosystem partly because
it has **no bypass environment variable**, and that is copied here on purpose.
I-6 — EPT parity plus asymmetry: an agent in good standing gets exactly what a
human of the same tier gets. Where a sibling cell silently demotes a tiered agent
to FREE because its EPT carries no tier, we do the opposite, and a test proves it.
"""
from __future__ import annotations
import logging
from dataclasses import dataclass
from enum import StrEnum
import httpx
from fastapi import Header, Request
from api.app.config import Settings
from api.app.errors import RepairPointer, passport_unresolvable
log = logging.getLogger(__name__)
class ActorType(StrEnum):
"""G3.7 — these three are the ONLY legal values.
A sibling service emits `actor_type: 'service'` into a telemetry ingest whose
Literal allows only human|agent|system, so every batch 422s and is dropped
with a single console warning. `service` is not spelled `service` here.
"""
human = "human"
agent = "agent"
system = "system"
@dataclass(frozen=True)
class Caller:
actor_type: ActorType
identity_id: str | None = None
passport: str | None = None
band: str | None = None
allowed_actions: tuple[str, ...] = ()
@property
def subject(self) -> str:
return self.identity_id or self.passport or "system"
# EI_CAPABILITY_MATRIX.v1 — velocity multipliers by integrity band.
BAND_MULTIPLIER: dict[str, float] = {
"platinum": 10.0,
"gold": 4.0,
"standard": 1.0,
"proven": 1.0,
"watch": 0.5,
"untrusted": 0.0, # read-only
# Eternitas began emitting this band on 2026-07-30 and it is not in the
# documented enum yet. Treating an unknown band as untrusted would lock out
# every freshly hatched agent; treating it as trusted would be a hole.
# Standard-with-no-bonus is the honest middle.
"unproven": 1.0,
}
async def resolve_passport(settings: Settings, passport: str) -> tuple[str, tuple[str, ...]]:
"""G3.6 — THE STATUS-CODE LAW.
404 and 400 REFUSE. 429 and 5xx retry with backoff, then REFUSE.
A sibling service maps 400 and 429 to "unreachable" and then soft-ALLOWS.
That is a live residual bypass, because 429 is trivially inducible at
100/min/IP: an attacker who wants the check skipped only has to make the
check rate-limit itself. There is no code path here where an unresolvable
passport is permitted to write.
"""
if not settings.eternitas_configured:
raise RepairPointer(
status_code=503,
code="trust_unavailable",
speak="We can't confirm helper IDs right now, so we didn't let that change through.",
machine_cause="eternitas is not configured; policy is fail-closed",
remediation_tool=None,
)
url = f"{settings.eternitas_base_url}/api/v1/trust/{passport}"
headers = {"X-API-Key": settings.eternitas_platform_api_key}
last_status = 0
for attempt in range(3):
async with httpx.AsyncClient(timeout=httpx.Timeout(8.0, connect=3.0)) as client:
try:
r = await client.get(url, headers=headers)
except httpx.RequestError as exc:
last_status = 599
log.warning("eternitas unreachable (attempt %s): %s", attempt + 1, exc)
continue
last_status = r.status_code
if r.status_code == 200:
body = r.json()
return body.get("band", "unproven"), tuple(body.get("allowed_actions", []))
if r.status_code in (400, 404):
# Malformed or not-issued. Refuse immediately — retrying cannot help
# and pretending it might is how a soft-allow gets written.
break
# 429 / 5xx: retry, then refuse. Never allow.
raise passport_unresolvable(passport, last_status)
async def get_caller(
request: Request,
authorization: str | None = Header(default=None),
x_service_token: str | None = Header(default=None),
) -> Caller:
settings: Settings = request.app.state.settings
# --- internal service caller (the Cloud portal) ------------------------
if x_service_token:
expected = settings.service_token
if not expected:
raise RepairPointer(
status_code=503,
code="service_auth_unconfigured",
speak="That connection isn't set up yet.",
machine_cause="SERVICE_TOKEN is unset; refusing to accept service calls",
remediation_tool=None,
)
# Constant-time compare, copied from the desktop control server's
# control-auth pattern rather than reinvented.
import hmac
if not hmac.compare_digest(x_service_token, expected):
raise RepairPointer(
status_code=401,
code="service_token_invalid",
speak="That connection isn't authorised.",
machine_cause="X-Service-Token did not match",
remediation_tool=None,
)
return Caller(actor_type=ActorType.system, identity_id="system")
if not authorization or not authorization.lower().startswith("bearer "):
raise RepairPointer(
status_code=401,
code="not_signed_in",
speak="You'll need to sign in first.",
machine_cause="no bearer token and no service token presented",
remediation_tool=None,
)
token = authorization.split(" ", 1)[1].strip()
# --- agent (Eternitas EPT) --------------------------------------------
passport = _unverified_claim(token, "passport") or _unverified_claim(token, "sub_passport")
if passport:
# ⚠️ SECURITY — trust is not authentication.
#
# A trust lookup answers "is this passport reputable?". It does NOT
# answer "does this caller actually hold this passport?". Skipping the
# second question is an authentication bypass: anyone who knows a
# passport number (they appear in logs, the lockbox and revocation
# messages) could present an UNSIGNED token naming it and be treated as
# that agent. Verified live 2026-08-13 — a forged `alg:none` token
# returned HTTP 200.
#
# ES256/JWKS verification of the EPT against Eternitas is not built yet
# (the G3.2/G9.1 verifier). Until it is, the agent path FAILS CLOSED in
# production — exactly as the human path below already does. This is not
# a downgrade of the "agents are citizens" design; it is refusing to
# seat a citizen whose ID we cannot yet check. It reopens automatically
# the moment `verify_ept_signature` exists and this gate consults it.
if settings.is_production and settings.require_verified_jwt:
raise RepairPointer(
status_code=503,
code="agent_signin_not_ready",
speak="Helper sign-in isn't switched on yet. Nothing you have is affected.",
machine_cause=(
"EPT signature verification (G3.2/G9.1) is not implemented; "
"refusing an unverified agent token in production. A trust "
"lookup proves reputation, not possession."
),
remediation_tool=None,
)
band, actions = await resolve_passport(settings, passport)
if band.lower() == "untrusted":
raise RepairPointer(
status_code=403,
code="agent_read_only",
speak="That helper can look, but it isn't allowed to make changes yet.",
machine_cause=f"passport {passport} band=untrusted is read-only",
remediation_tool=None,
)
return Caller(
actor_type=ActorType.agent,
passport=passport,
band=band,
allowed_actions=actions,
)
# --- human (account-server RS256) -------------------------------------
if settings.is_production and settings.require_verified_jwt:
# I-8, applied to ourselves. G3.2's JWKS verifier is not written yet, and
# an unverified JWT is an authentication bypass rather than a shortcut.
# Refusing is the only honest answer until the verifier exists.
raise RepairPointer(
status_code=503,
code="human_signin_not_ready",
speak="Signing in isn't switched on yet. Nothing you have is affected.",
machine_cause=(
"JWKS verification (G3.2) is not implemented; refusing to accept "
"an unverified human token in production"
),
remediation_tool=None,
)
identity_id = _unverified_claim(token, "windy_identity_id") or _unverified_claim(token, "sub")
if not identity_id:
raise RepairPointer(
status_code=401,
code="token_unrecognised",
speak="We couldn't read that sign-in. Try signing in again.",
machine_cause="token carried neither a passport nor an identity claim",
remediation_tool=None,
)
return Caller(actor_type=ActorType.human, identity_id=identity_id)
def _unverified_claim(token: str, claim: str) -> str | None:
"""Read a claim WITHOUT verifying the signature.
Used only to decide which verifier a token belongs to. Every path that acts
on the result re-establishes trust independently: an agent's authority comes
from a live Eternitas trust lookup, never from the token's own assertions.
⚠️ Full RS256/ES256 JWKS verification for the human path lands in G3.2's
verifier and MUST be in place before `api.windygit.com` accepts a human
token from outside. Until then the human path is reachable only from inside
the tunnel, and `settings.require_verified_jwt` refuses it in production.
"""
import base64
import json
try:
payload = token.split(".")[1]
payload += "=" * (-len(payload) % 4)
return json.loads(base64.urlsafe_b64decode(payload)).get(claim)
except Exception: # noqa: BLE001
return None