flex-auth/SCOPE.md
tegwick c0d0d92e9f Make the layer declaration a boundary, and review the boundaries it implies.
INTENT.md pinned standard_version: "0.7" in the frontmatter §11 requires. That
conflated two things the standard separates itself: assent "records assent to a
BOUNDARY, given at the version named. It is not assent to the current text."
flex-auth is Engine/PDP at v0.6, v0.7, v0.8 and after; the role does not change
when the text is amended. The field was also decorative — parsed into
Declaration.StandardVersion and never validated — so the version was load-bearing
only via a test asserting it equalled 0.7.

That test is inverted rather than deleted: internal/layer now rejects a version
pin in the declaration and requires conformance_record to name a file that
exists. Version-scoped state moves to docs/conformance/security-layer-conformance.md,
a derived artifact carrying what it derives from and the version derived at, as
§11 requires of derived artifacts.

SCOPE.md: gap assessment replaces "conforming with one declared gap" with three
gaps, each with an owner and a route. G2 is new — flex-auth declares no emission
guarantee where §11 requires one of every §4 source of evidence. It is recorded
as a gap rather than as conformance because the flattering reading, that
audit-core is the source and flex-auth merely produces, has been asserted by
nobody but flex-auth. Also corrects the stance register from two rows to five.

Fixing one line meant reading what the declaration asserts, and a boundary is
only half held here. docs/conformance/boundaries-review.md checks the other
halves across twelve counterparts and finds four security-relevant repositories
with no layer declaration at all — including key-cape, the identity source whose
claims flex-auth consumes as normative input. That boundary is asserted from one
side only. Recorded as unstated, never as agreed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 28468@bnt-lap001
Assistant-Session: c76569b2-6056-4dad-aea4-49cd7a018f5d
2026-09-21 00:11:56 +02:00

347 lines
18 KiB
Markdown

# flex-auth Scope
## One-liner
flex-auth is a policy-as-code authorization registry and decision engine for
NetKingdom-aligned systems. Under the NetKingdom Security Layer Model
(declared 2026-08-29) it is **Engine / PDP** — the estate's only policy decision
point.
## Core Idea
flex-auth turns verified identity claims, resource manifests, relationship
facts, policy packages, and request context into inspectable authorization
decisions. It is intentionally separate from identity providers and from the
applications that enforce decisions.
The product should start as a useful standalone local authorization system,
then grow into delegated mode where Topaz, OpenFGA, SpiceDB, OPA, Cedar,
Keycloak Authorization Services, Entra, Graph, SCIM, LDAP, and similar systems
can be coordinated behind a stable flex-auth API.
## In Scope
- Protected-system registration.
- Resource namespaces, resource hierarchy, and action vocabularies.
- Subject, group, team, tenant, role, and service-account records needed for
authorization.
- Relationship facts and inherited access.
- Versioned policy packages with metadata, fixtures, validation, tests, and
activation state.
- Deterministic check, batch-check, list-allowed, explain, and decision-record
capabilities.
- Local decision logs with policy version and provenance.
- CLI and service entry points for local development and integration tests.
- Markitect as the first concrete protected-system consumer.
- Delegated PDP and directory adapter boundaries after the standalone core is
stable.
- The **decision-record schema**, published as flex-auth's own contract
(Security Layer Model §17).
- The **canonical request digest** as the published replay test for consumers
(§6.4.2) — `binding.submitted_request_digest`, over the request as sent.
`request_digest` is over the *enriched* request and is flex-auth's own
audit-replay identity, not a consumer check (`FLEX-DEC-2026-012`).
- **Decision lifetimes** on every allow, and **revocation visibility deadlines
stated per input class** (§9.7).
- **Registry-snapshot provenance**, so a decision that turned on registry
content is replayable from its own record (§6.1, §9.7.2).
## Out of Scope
- Acting as an identity provider.
- Replacing key-cape, NetKingdom SSO, Keycloak, Entra, or other identity
systems.
- Owning login, MFA, PKCE, token issuance, token lifecycle, or canonical
identity profile behavior.
- Embedding protected-system enforcement logic inside flex-auth.
- Making Markitect-specific assumptions part of the generic authorization
model.
- Forcing a single PDP backend or directory provider.
- Hiding policy behavior behind opaque administrative toggles.
- Expressing fail-open. When flex-auth is unreachable there is no evaluator in
the path; the residue is the consumer's declared stance (§9.3).
- Owning enforcement. A PEP's obligations under §6.4 are the consumer's,
including its published stance map.
- Owning the approval object, its storage, or its lifecycle — that is
`approval-engine`'s (§9.4). flex-auth validates approvals as input claims and
never mutates them.
- Owning evidence of authentication or assurance. flex-auth consumes assurance
claims and never redefines them; it owns evidence of the decision
(`FLEX-DEC-2026-002`, recorded as declined in §13).
- Authoring security doctrine. gate-house owns invariants, authority ceilings,
operating modes, and the authority context flex-auth consumes as claims.
## Relevant When
- A protected system needs resource-specific authorization decisions.
- A team needs policy-as-code packages that can be reviewed, tested, versioned,
and explained.
- Authorization must combine identity claims, groups, roles, tenants,
relationships, context, and resource metadata.
- A local development workflow needs deterministic authorization behavior
before enterprise PDP integrations exist.
- Markitect resources need central authorization decisions and durable audit
records.
- A delegated backend needs to be introduced without changing the
protected-system-facing API.
## Not Relevant When
- The task is only about authentication, login, token issuance, or MFA.
- The desired outcome can be handled by coarse identity-provider roles alone.
- The protected system only needs local development labels and no central
policy administration.
- A team is choosing identity-provider configuration rather than
resource-specific authorization.
## Current State
The standalone core is implemented and **all seven baseline workplans
(`FLEX-WP-0001` through `FLEX-WP-0007`) are complete.** The repository carries
the intent baseline, authorization landscape research, ADR set, and a working
Go service (`cmd/flex-auth`) with `validate`, `load-registry`, `serve`, and
`POST /v1/check` plus registry, policy, decision, audit, Markitect, and
delegated-adapter internals. The standalone policy-as-code core (`FLEX-WP-0002`),
Markitect consumer integration (`FLEX-WP-0003`, manifest ingest, decisions, and
fixtures), and the delegated PDP/directory adapter shapes (`FLEX-WP-0004`,
Topaz/OpenFGA/OPA/Cedar/Keycloak/Entra tradeoffs documented with at least one
controlled adapter shape) all landed in May 2026.
Standalone decisions include a structured `binding` containing the normalized
subject, action, resource, context, and a full request digest. This makes the
execute-time decision contract inspectable without parsing reason or diagnostic
prose. Durable human approval lifecycle, supersession, and authenticated
multi-approver evidence are `approval-engine` responsibilities; the proposed
cross-system object is documented in
`schemas/action_authorization.schema.json` and is not yet a deployed
`approval-engine` endpoint.
**secrets-engine is a shipped consumer as of 2026-09-06** (`FLEX-WP-0021`):
`secrets-engine.catalog-lane.lifecycle` v2 over `resource.type:
secret-catalog-lane`, twelve actions delivered by secrets-engine rather than
inferred, `destroy` gated on a dual-control approval claim, and a dedicated
`flex-auth-secrets-engine` pin at
`http://flex-auth-secrets-engine.flex-auth.svc.cluster.local.:8080`
trailing dot required, in-cluster only — with `callerAuth.mode: warn`. Adoption is not complete: secrets-engine is a CLI
rather than a workload, and the pin's default-deny ingress admits a pod, so an
operator-run access path is still undecided.
v1 of that package **had no tenant rule and allowed a foreign tenant**; v2
supersedes rather than amends it, because a fail-open correction must be visible
to a consumer as a version change (`FLEX-DEC-2026-008`). The sweep that finding
prompted shows `tenant-engine` unscoped on tenant as well, carried by
`FLEX-WP-0022`.
The **first shipped protected-system consumer is ops-warden**: its opt-in
pre-sign gate calls `POST /v1/check` for `resource.type: ssh-certificate`,
`action: sign` decisions (`examples/ops-warden/`, policy package, allow/deny
fixtures, and tests). `FLEX-WP-0006` published that gate and `FLEX-WP-0007`
deployed flex-auth as a reachable production runtime for it. The joint
OpenBao-backed smoke is verified (2026-06-29: vault-backed allow recorded
`decision:032b096c433ad80c`; TTL-over-max denied `ttl_out_of_bounds` by
flex-auth before OpenBao). `FLEX-WP-0016` then shipped an independently rollable
in-cluster ops-warden pin and moved it to `callerAuth.mode: enforce` once
ops-warden adopted a calling identity (`WARDEN-WP-0031`): anonymous `/v1/check`
is 401, a token bound to another protected system is 403.
Production `policy.enabled` is **deliberately off, and is not coming back as a
boolean.** ops-warden's `ADR-0009` (2026-08-22, superseding `ADR-0006`) settles it:
`policy.enabled` is one repo-wide flag, and with `fail_closed: true` it makes
flex-auth a hard dependency of every `warden sign` — including the SSH
certificates the ops-bridge tunnels depend on, one of which carries the policy
call itself. `ADR-0009` retires `policy.enabled` and the global `policy.fail_closed`
outright and replaces them with a total per-zone stance map in the consumer PEP
(`ops-warden/pep-stance.yaml`): open for `z0`-`z2` and `unknown`, closed for
`z3-critical`. That map is the reference shape for the Security Layer Model's
§6.4 obligation 3. It is no longer the only published one: `secrets-engine`
published `pep-stance.yaml` (total over catalog stage plus `unknown`, runtime-read
and test-pinned), so §13.1's register grew past the single row the standard
recorded as itself the finding. It now carries **five**`ops-warden`,
`user-engine`, `tenant-engine`, `secrets-engine`, and `ops-mason` with no map at
all — two of them marked non-conformant. flex-auth reviewed the pair in
`docs/stance-register-review.md` — the first exercise of the aggregate-divergence
capability it claimed on 2026-08-29 — and found that the two maps take opposite
stances on `unknown` (`fail_open` vs `fail_closed`) and scope on different axes
(security-zone vs catalog-stage), which makes the rows conformant but not
comparable.
flex-auth reviewed the zone model as the PDP on 2026-08-19 and holds one
boundary: zone **membership** compiles into
the registry snapshot flex-auth already consumes (no schema change needed), while
per-zone **stance** belongs in the versioned policy package, because registry
content is absent from decision provenance. **Fail-open is not expressible by a
PDP at all** and stays a consumer-side setting. The enforcing pins stay
enforcing regardless — an enforcing pin with no enforcing consumer still turns
the anonymous path into a 401 rather than a decision.
**Layer position (declared 2026-08-29).** flex-auth is **Engine / PDP** under
the NetKingdom Security Layer Model, declared machine-readably in `INTENT.md`
frontmatter per §11. That declaration deliberately carries **no standard
version**: a layer is a boundary and does not change when the text is amended,
which is the distinction the standard itself draws when it says assent is *"to a
BOUNDARY, given at the version named. It is not assent to the current text."* The
absence is enforced by test in `internal/layer`, and raised to `gate-house` as
`FLEX-WP-0030` B4 in case §11 should say so generally.
Version-scoped state therefore lives in
`docs/conformance/security-layer-conformance.md` — a derived artifact that names
what it derives from and the version it was derived at, as §11 requires. It
carries the assent history (v0.4 `FLEX-DEC-2026-002`, v0.6 `FLEX-DEC-2026-003`,
v0.8 `FLEX-DEC-2026-011`) and the gap table below.
Findings of flex-auth's adopted into the standard: §9.3's two-owner split for
degraded mode, §6.4.2 scoped to the decision's own binding with the canonical
request digest as its mechanical test and negative caching ruled explicitly,
§9.7.2 split by role, §17 moving the decision-record schema to flex-auth, and
four more in the v0.8 round (`FLEX-DEC-2026-011`), of which F1 — that a decision
must be *attributable* to flex-auth and that a digest comparison does not
discharge it — was the finding of the round.
Conformance state is **conforming with three declared gaps**, each with an owner
and a route rather than a sentence:
| # | Gap | Owner | Route |
| --- | --- | --- | --- |
| G1 | Registry-snapshot digest absent from decision provenance (§9.7.2 conformance prerequisite) | `flex-auth` | `FLEX-WP-0019` |
| G2 | No emission guarantee declared, where §11 requires one of every §4 source of evidence | `gate-house` to rule, then `flex-auth` | `FLEX-WP-0030` B3 |
| G3 | Published stance-register review stale — written at two register rows, §13.1 now carries five | `flex-auth` | `FLEX-WP-0029` |
G2 is new and is recorded as a gap rather than as conformance on purpose: whether
flex-auth is a §4 *source of evidence* or only the producer of an artifact
`audit-core` sources has been asserted by nobody but flex-auth, and §11 says a
source declaring no emission guarantee is not conforming. The conservative entry
is the honest one until the boundary is ruled.
**Boundaries review (2026-09-21).** `FLEX-WP-0030` reviews flex-auth's boundary
against every security-relevant counterpart and raises five items: inconsistent
`layer:` casing across the estate against §11's mechanical-checkability claim
(B1); four security-relevant repositories with no layer declaration at all —
`gate-house`, `key-cape`, `ops-mason`, `net-kingdom` — of which `key-cape` is the
identity source whose claims flex-auth consumes as normative input, making that
boundary asserted from one side only (B2); G2 above (B3); B4; and canon naming
`access-engine` while the repository still answers to `flex-auth` (B5).
State Hub integration is present through:
- `.custodian-brief.md`
- `state_hub_workstream_id` fields in workplans
- `state_hub_task_id` fields in task blocks
- active State Hub dispatch for the `flex-auth` repo
## How It Fits
flex-auth sits between identity and protected systems:
```text
key-cape / NetKingdom SSO
-> verified identity claims
-> flex-auth authorization registry and policy decisions
-> protected systems such as Markitect
```
Identity systems establish who the caller is and provide coarse claims.
flex-auth decides what that caller can do to a specific resource under a
specific context. Protected systems enforce the decision and keep their own
local diagnostics.
## Terminology
- **Subject:** Human, service account, group, team, tenant, role, or emergency
principal.
- **Resource:** Protected object registered by a system.
- **Namespace:** Resource type and ownership boundary.
- **Action:** Operation requested against a resource.
- **Context:** Request, environment, assurance, workflow, or runtime
attributes.
- **Policy package:** Versioned policy-as-code bundle with metadata and tests.
- **Relationship fact:** Subject-resource or resource-resource relation.
- **Decision:** Allow, deny, redact, audit-only, or not-applicable outcome with
reason and provenance.
- **PDP:** Policy decision point, either flex-auth standalone logic or a
delegated backend.
- **PEP:** Policy enforcement point in the protected system.
## Related / Overlapping
- key-cape / NetKingdom SSO: identity source and coarse claims provider;
flex-auth consumes the **NetKingdom IAM Profile**
(`~/net-kingdom/canon/standards/iam-profile_v0.2.md`).
- ops-warden: first **shipped** protected-system consumer. Its opt-in
pre-sign gate calls flex-auth for `ssh-certificate` / `sign` decisions
before issuing a short-lived SSH certificate (`FLEX-WP-0006`,
`FLEX-WP-0007`). ops-warden owns the SSH CA, OpenBao signing, and actor
inventory; flex-auth owns the policy decision. ops-warden's routing
charter names flex-auth as the owner of every "may I perform action X?"
question.
- Markitect: first **knowledge-pipeline** consumer. Integration is complete on
the flex-auth side (`FLEX-WP-0003` — resource-manifest ingest, Markitect-
compatible decisions, and fixtures); a live Markitect runtime calling the gate
in production is the next consumer milestone after ops-warden.
- Topaz: aligned evaluator. Per ADR-003 the standalone core is shaped
to match Topaz's Rego + directory model from day one; the Topaz
adapter in `FLEX-WP-0004` is therefore a small step rather than a
conversion.
- OpenFGA and SpiceDB: candidate relationship authorization backends.
- OPA and Cedar: candidate rule and typed-policy engines.
- Keycloak Authorization Services: adapter path for Keycloak-centric
deployments. Default architecture is "Keycloak as SSO only,
flex-auth owns authorization"; Keycloak AuthZ is one optional
delegated PDP.
- Entra, Graph, SCIM, LDAP, and Keycloak APIs: directory and group resolver
sources.
## Disjoint From
- **ops-warden** is a flex-auth *consumer*, not an overlap (see Related /
Overlapping). The two remain disjoint on **identity surface**: ops-warden
issues SSH certificates for ops actors (`adm`/`agt`/`atm`) and is not a
resource-policy engine; flex-auth decides whether a given sign request is
allowed and never issues certificates. The once-hypothetical flow of
surfacing an `agt` actor as a flex-auth subject is now realized through
the signing policy gate.
- **ops-bridge** owns SSH reverse-tunnel connectivity and explicitly
disclaims being a credential authority or policy engine. No overlap.
## Provided Capabilities
```capability
type: orientation
title: Agent orientation
status: current
description: Repo intent, boundaries, research, workplan map, and State Hub agent brief are available for new coding agents.
```
```capability
type: workplan
title: Standalone policy core plan
status: current
description: Workplan for schemas, local registry, policy packages, check APIs, list/explain, decision log, CLI/service skeleton, tests, and examples.
```
```capability
type: workplan
title: Markitect consumer integration plan
status: current
description: Workplan for Markitect resource namespace, manifest import, action vocabulary, check fixtures, contract tests, and integration flow.
```
```capability
type: workplan
title: Delegated adapter plan
status: current
description: Workplan for Topaz evaluation, relationship PDP adapters, rule PDP adapters, Keycloak AuthZ path, directory resolvers, and operations docs.
```
```capability
type: contract
title: Decision-record contract
status: current
description: Published decision-record schema, the submitted-request digest as the replay test for consumers, request-enrichment precedence, and decision provenance — flex-auth's own contract under the Security Layer Model section 17.
```
```capability
type: orientation
title: Layer declaration
status: current
description: Machine-readable Engine/PDP boundary declaration in INTENT.md frontmatter, carrying no standard version by design, with version-stamped conformance state and three declared gaps in docs/conformance/security-layer-conformance.md, per Security Layer Model section 11.
```