feat: complete and prove the authorization chain end to end

Implements step 2 (access-engine POST /v1/check) and wires the whole
GH-DEC-2026-003 sequence together, then proves it against a live throwaway
OpenBao rather than only unit-level fakes.

- decision_check.check_decision performs the PDP call; an unreachable or
  non-200 PDP raises, since silence is never permission.
- approval_consume.authorize_action coordinates steps 1 and 2 and returns an
  AuthorizedAction. Both steps build the same CheckRequest via a shared
  _expected_request, since two descriptions of the action cannot produce
  corresponding digests.
- apply_unreachable_engine_stance takes authorized=. The published map
  defines fail_closed as no side effect WITHOUT a durable decision record,
  so holding a validated one means the residue does not apply. Not a bypass:
  both steps must have succeeded and CAS consume still precedes OpenBao.
  Unconfigured still returns None and fails closed.

The end-to-end test caught one more instance of the cross-vocabulary bug: a
leftover comparison of the claim's binding.action against ours. The claim
says secrets.kv.destroy where we say destroy, so it would have failed against
every real claim. Removed; the tie is pdp_digest.

Integration coverage asserts PIP-then-PDP ordering, that consume is the last
step before the backend, and that an unreachable PDP, denied decision,
invalid claim, missing pdp_digest, consume conflict and action mismatch each
stop before OpenBao. 284 tests pass; production still fails closed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01M65ovP3eiiPHubibvWs9mD

Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 393550@bnt-lap001
Assistant-Session: 4bb359f9-1f12-4410-9e76-079cf23c82e4
This commit is contained in:
tegwick 2026-09-06 14:56:02 +02:00
parent e8144f315c
commit f62d3fe789
9 changed files with 675 additions and 44 deletions

View file

@ -19,7 +19,12 @@ from urllib.error import HTTPError, URLError
from urllib.request import Request, urlopen
from secrets_engine.approval_claim import validate_approval_claim
from secrets_engine.authorization import build_action_request, request_digest
from secrets_engine.decision_check import check_decision
from secrets_engine.authorization import (
build_action_request,
request_digest,
validate_decision_envelope,
)
from secrets_engine.errors import DecisionError
from secrets_engine.openbao import read_strict_token_file
from secrets_engine.pep_stance import demo_exception_enabled
@ -37,6 +42,26 @@ class ConsumeBinding:
decision_id: str = ""
@dataclass(frozen=True)
class AuthorizedAction:
"""A validated claim and decision for one exact proposed action.
Holding this is not authority to act: GH-DEC-2026-003 still requires a
successful CAS consume before the OpenBao call.
"""
binding: ConsumeBinding
decision_id: str
expires_at: str
def as_evidence(self) -> dict[str, object]:
return {
"authorization_decision_id": self.decision_id,
"authorization_expires_at": self.expires_at,
"request_digest": self.binding.request_digest,
}
@dataclass(frozen=True)
class ConsumedApproval:
"""Non-secret confirmation that consume succeeded for this request."""
@ -94,6 +119,44 @@ def _request_purpose(entry: Any) -> str:
return ""
def _expected_request(
cfg: Any,
entry: Any,
action: str,
*,
fields: tuple[str, ...] = (),
policy_targets: tuple[str, ...] = (),
auth_targets: tuple[str, ...] = (),
) -> dict[str, Any]:
"""Build the exact CheckRequest both steps must agree on.
Steps 1 and 2 must describe the same proposed action or the digests cannot
correspond, so neither builds its own.
"""
subject_id = str(getattr(cfg, "authorization_subject_id", "") or "")
subject_type = str(getattr(cfg, "authorization_subject_type", "") or "")
if not subject_id or not subject_type:
raise DecisionError(
"authorization join requires SECRETS_ENGINE_AUTHORIZATION_SUBJECT_ID "
"and _SUBJECT_TYPE; the PEP must not assert an unnamed subject"
)
purpose = _request_purpose(entry)
if not purpose:
raise DecisionError(
"authorization join requires a declared approval/consumer purpose"
)
return build_action_request(
entry,
action,
subject_id=subject_id,
subject_type=subject_type,
purpose=purpose,
fields=fields,
policy_targets=policy_targets,
auth_targets=auth_targets,
)
def fetch_approval_claim(
*,
base_url: str,
@ -147,18 +210,12 @@ def resolve_consume_binding(
auth_targets: tuple[str, ...] = (),
opener: Callable[..., Any] | None = None,
) -> ConsumeBinding | None:
"""Join the proposed action to a served approval-claim (step 1 of GH-DEC-2026-003).
"""Join the proposed action to a served approval-claim (step 1).
Returns None only when no serving path is configured at all, keeping
production fail-closed exactly as it was before the join existed. Anything
configured-but-wrong raises: a half-configured PEP must not look like an
unconfigured one.
Step 2 (the flex-auth DecisionEnvelope from POST /v1/check) is validated by
``authorization.validate_decision_envelope``. No PDP is reachable for this
consumer yet -- flex-auth runs per-consumer cluster-local pins and
``flex-auth-secrets-engine`` has not been created -- so that call is not
wired here and production stays closed at the stance gate regardless.
"""
base_url = str(getattr(cfg, "approval_url", "") or "")
token_file = getattr(cfg, "approval_token_file", None)
@ -166,29 +223,11 @@ def resolve_consume_binding(
if not base_url or not token_file or not authorization_id:
return None
subject_id = str(getattr(cfg, "authorization_subject_id", "") or "")
subject_type = str(getattr(cfg, "authorization_subject_type", "") or "")
if not subject_id or not subject_type:
raise DecisionError(
"authorization join requires SECRETS_ENGINE_AUTHORIZATION_SUBJECT_ID "
"and _SUBJECT_TYPE; the PEP must not assert an unnamed subject"
)
purpose = _request_purpose(entry)
if not purpose:
raise DecisionError(
"authorization join requires a declared approval/consumer purpose"
)
expected_request = build_action_request(
entry,
action,
subject_id=subject_id,
subject_type=subject_type,
purpose=purpose,
fields=fields,
policy_targets=policy_targets,
auth_targets=auth_targets,
expected_request = _expected_request(
cfg, entry, action,
fields=fields, policy_targets=policy_targets, auth_targets=auth_targets,
)
# Two different digests over the same proposed action, by contract; they are
# never compared to each other.
#
@ -215,9 +254,11 @@ def resolve_consume_binding(
approval_id=authorization_id,
expected_pdp_digest=pdp_digest,
)
binding = claim.get("binding") or {}
if isinstance(binding, dict) and binding.get("action") not in (None, action):
raise DecisionError("approval claim does not bind this action")
# No action comparison here. The claim's binding.action is approval-engine
# vocabulary ("secrets.kv.destroy") and ours is the catalog's ("destroy");
# comparing them would fail against every real claim, which is the same
# cross-vocabulary mistake the native digest made. The tie to this exact
# action is pdp_digest, checked above.
return ConsumeBinding(
approval_id=authorization_id,
request_digest=pdp_digest,
@ -225,6 +266,80 @@ def resolve_consume_binding(
)
def authorize_action(
cfg: Any,
entry: Any,
action: str,
decision: Any = None,
*,
fields: tuple[str, ...] = (),
policy_targets: tuple[str, ...] = (),
auth_targets: tuple[str, ...] = (),
opener: Callable[..., Any] | None = None,
pdp_opener: Callable[..., Any] | None = None,
) -> AuthorizedAction | None:
"""Run steps 1 and 2 for one proposed action, or return None if unserved.
Step 1 validates the approval-claim; step 2 obtains and validates the
flex-auth DecisionEnvelope. Returning None means no serving path is
configured at all, which leaves production fail-closed. A configured but
failing path raises: a partial deployment must not read as an absent one.
"""
binding = resolve_consume_binding(
cfg, entry, action, decision,
fields=fields,
policy_targets=policy_targets,
auth_targets=auth_targets,
opener=opener,
)
if binding is None:
return None
pdp_url = str(getattr(cfg, "pdp_url", "") or "")
pdp_token = getattr(cfg, "pdp_token_file", None)
if not pdp_url or not pdp_token:
raise DecisionError(
"production action requires an access-engine decision; "
"SECRETS_ENGINE_PDP_URL / _PDP_TOKEN_FILE are unset"
)
package = str(getattr(cfg, "authorization_policy_package", "") or "")
version = str(getattr(cfg, "authorization_policy_version", "") or "")
if not package or not version:
raise DecisionError(
"authorization join requires an explicitly configured policy "
"package/version pin; the reserved coordinate is a reservation, "
"not a publication, and must not be used as a default"
)
expected_request = _expected_request(
cfg, entry, action,
fields=fields, policy_targets=policy_targets, auth_targets=auth_targets,
)
envelope = check_decision(
base_url=pdp_url,
token_file=Path(pdp_token),
request=expected_request,
opener=pdp_opener or urlopen,
)
validated = validate_decision_envelope(
envelope,
expected_request,
accepted_policy_packages={package},
accepted_policy_versions={version},
)
if validated.action != action:
raise DecisionError("access-engine decision does not bind this action")
return AuthorizedAction(
binding=ConsumeBinding(
approval_id=binding.approval_id,
request_digest=binding.request_digest,
decision_id=validated.decision_id,
),
decision_id=validated.decision_id,
expires_at=validated.expires_at,
)
def consume_approval(
*,
base_url: str,

View file

@ -35,6 +35,7 @@ from types import SimpleNamespace
from secrets_engine import __version__
from secrets_engine.apply import apply_plan
from secrets_engine.approval_consume import (
authorize_action,
require_production_consume,
resolve_consume_binding,
)
@ -122,9 +123,19 @@ def _require_lane_approval(
consume path. Build/test ``fail_open`` still requires the existing
lane-approval check a tracked gap until SECRETS-WP-0008-T02.
"""
stance = apply_unreachable_engine_stance(cfg, entry, action or "unknown")
# Steps 1 and 2 first: a validated claim and decision are what make the
# unreachable-engine residue inapplicable. Absent configuration returns
# None and production stays closed exactly as before.
authorization = authorize_action(
cfg, entry, action or "unknown", None, fields=fields
)
stance = apply_unreachable_engine_stance(
cfg, entry, action or "unknown", authorized=authorization is not None
)
if evidence is not None:
evidence.mark_stance(stance)
if authorization is not None:
evidence.detail.update(authorization.as_evidence())
decision = None
if entry.approval_required():
decision = resolve_decision(
@ -138,9 +149,7 @@ def _require_lane_approval(
require_production_consume(
cfg,
entry,
binding=resolve_consume_binding(
cfg, entry, action or "unknown", decision, fields=fields
),
binding=authorization.binding if authorization is not None else None,
evidence=evidence,
)
return decision

View file

@ -49,6 +49,8 @@ class Config:
authorization_policy_package: str = ""
authorization_policy_version: str = ""
authorization_min_approvals: int = 1
pdp_url: str = ""
pdp_token_file: Path | None = None
@classmethod
def load(cls) -> "Config":
@ -56,6 +58,7 @@ class Config:
token_file = os.environ.get("SECRETS_ENGINE_APPROVAL_TOKEN_FILE", "")
keycape_secret = os.environ.get("SECRETS_ENGINE_KEYCAPE_CLIENT_SECRET_FILE", "")
jwt_login = os.environ.get("SECRETS_ENGINE_OPENBAO_JWT_LOGIN", "")
pdp_token = os.environ.get("SECRETS_ENGINE_PDP_TOKEN_FILE", "")
return cls(
catalog_dir=Path(os.environ.get("SECRETS_ENGINE_CATALOG", root / "catalog")),
policy_dir=Path(os.environ.get("SECRETS_ENGINE_POLICIES", root / "policies")),
@ -86,4 +89,6 @@ class Config:
authorization_min_approvals=_positive_int(
os.environ.get("SECRETS_ENGINE_AUTHORIZATION_MIN_APPROVALS", "")
),
pdp_url=os.environ.get("SECRETS_ENGINE_PDP_URL", ""),
pdp_token_file=Path(pdp_token) if pdp_token else None,
)

View file

@ -0,0 +1,73 @@
"""access-engine (flex-auth) Check client — step 2 of GH-DEC-2026-003.
This engine consumes a decision; it never renders one. The DecisionEnvelope
returned here is validated by ``authorization.validate_decision_envelope``
against the exact proposed action before it can satisfy the production stance.
No estate-wide PDP exists by design: flex-auth runs per-consumer cluster-local
pins, so the address is per-deployment configuration and its absence fails
production closed.
"""
from __future__ import annotations
import json
from pathlib import Path
from typing import Any, Callable
from urllib.error import HTTPError, URLError
from urllib.request import Request, urlopen
from secrets_engine.errors import DecisionError
from secrets_engine.openbao import read_strict_token_file
_MAX_BODY = 512 * 1024
def _status_message(status: int) -> str:
if status in (401, 403):
return "access-engine refused the caller"
if status == 404:
return "access-engine has no such endpoint"
if status == 503:
return "access-engine is unavailable"
return f"access-engine returned HTTP {status}"
def check_decision(
*,
base_url: str,
token_file: Path,
request: dict[str, Any],
timeout_seconds: float = 3,
opener: Callable[..., Any] = urlopen,
) -> dict[str, Any]:
"""POST /v1/check. Any non-200, non-JSON, or transport failure fails closed.
Silence is never permission: an unreachable PDP raises rather than
returning a permissive default.
"""
if not base_url or not base_url.startswith(("http://", "https://")):
raise DecisionError("access-engine check URL is missing or invalid")
token = read_strict_token_file(Path(token_file), purpose="access-engine credential")
encoded = json.dumps(request).encode("utf-8")
http_request = Request(
base_url.rstrip("/") + "/v1/check",
data=encoded,
method="POST",
)
http_request.add_header("Authorization", f"Bearer {token}")
http_request.add_header("Content-Type", "application/json")
http_request.add_header("Accept", "application/json")
try:
with opener(http_request, timeout=timeout_seconds) as response:
if getattr(response, "status", 200) != 200:
raise DecisionError("access-engine check did not return a decision")
payload = json.loads(response.read(_MAX_BODY).decode("utf-8"))
except HTTPError as e:
raise DecisionError(f"access-engine check refused: {_status_message(e.code)}") from e
except URLError as e:
raise DecisionError("access-engine is unreachable for check") from e
except json.JSONDecodeError as e:
raise DecisionError("access-engine check returned a non-JSON body") from e
if not isinstance(payload, dict):
raise DecisionError("access-engine check returned a non-object body")
return payload

View file

@ -36,12 +36,14 @@ class StanceApplication:
action: str
demo_exception: bool = False
decision_id: str = ""
authorized: bool = False
def as_evidence(self) -> dict[str, object]:
payload: dict[str, object] = {
"stance_stage": self.stage,
"stance_failure_mode": self.failure_mode,
"stance_demo_exception": self.demo_exception,
"stance_authorized": self.authorized,
}
if self.decision_id:
payload["stance_decision_id"] = self.decision_id
@ -111,12 +113,20 @@ def apply_unreachable_engine_stance(
action: str,
*,
stance_map: PepStanceMap | None = None,
authorized: bool = False,
) -> StanceApplication:
"""Apply the published unreachable-engine residue for a live action.
``fail_closed`` without the demo exception raises ``DecisionError`` carrying
named stance fields. ``fail_open`` is the documented residue: continue to
the existing lane-approval check, which is itself a gap until T02.
``fail_closed`` is the *unreachable-engine* residue, not a blanket ban: the
published map defines it as no protected side effect without a durable
access-engine decision record. ``authorized=True`` means the caller already
obtained and validated that record for this exact action, so the engine was
reachable and the residue does not apply. It is never a bypass -- the caller
must have completed steps 1 and 2, and GH-DEC-2026-003 still requires a
successful CAS consume before any OpenBao call.
Without such a record, ``fail_closed`` raises ``DecisionError`` carrying
named stance fields. ``fail_open`` is the documented residue for build/test.
"""
loaded = stance_map or load_pep_stance()
stage, mode = loaded.for_stage(getattr(entry, "stage", "unknown"))
@ -126,8 +136,9 @@ def apply_unreachable_engine_stance(
failure_mode=mode,
action=action or "unknown",
demo_exception=bool(demo and mode == "fail_closed"),
authorized=bool(authorized),
)
if mode == "fail_closed" and not demo:
if mode == "fail_closed" and not demo and not authorized:
raise DecisionError(
f"production action '{applied.action}' requires a durable "
"access-engine decision record; live production remains disabled",