flex-auth/workplans/FLEX-WP-0023-operator-caller-access-path.md

235 lines
10 KiB
Markdown
Raw Normal View History

docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
---
id: FLEX-WP-0023
type: workplan
title: "Operator caller access path and caller identity in the decision record"
domain: infotech
repo: flex-auth
status: active
docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
owner: claude
topic_slug: netkingdom
planning_priority: P1
planning_order: 230
depends_on_workplans:
- FLEX-WP-0021
related_workplans:
- FLEX-WP-0016
- SECRETS-WP-0009
created: "2026-09-06"
updated: "2026-09-06"
state_hub_workstream_id: "ad011f92-786c-51ad-b3f6-c06ad77e7af7"
docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
---
# FLEX-WP-0023 — Operator caller access path and caller identity in the decision record
Opened by `glas-harness` (`GLAS-WP-0015`, 2026-09-06), which asked flex-auth to
keep the access-path gate live work despite `FLEX-WP-0021-T05` closing, and to
return "a supported owner access path and authenticated caller binding/lifetime
with positive and negative tests". They ruled out Service DNS and a permanent
operator token by name. Both refusals are correct.
Design: [`../docs/operator-caller-access-path.md`](../docs/operator-caller-access-path.md).
Contract gap: `FLEX-DEC-2026-009`.
## The finding that reframes the task
For an operator caller the pin's default-deny `NetworkPolicy` is not a partial
control, it is **silent**: `kubectl port-forward` is proxied to the pod's own
loopback and no policy selector is consulted. So with `callerAuth.mode: warn`,
the operator path is unauthenticated *and* unfiltered, and the only reason it is
not reachable today is that nobody has forwarded the port.
`warn` was a safe migration state for `ops-warden` because a workload pin still
admitted exactly one pod. Here the warn window protects nothing while it runs.
**`enforce` is the deliverable, not the follow-up.**
## 1. Create the ServiceAccount the deployed binding already names
```task
id: FLEX-WP-0023-T01
status: done
docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
priority: high
state_hub_task_id: "10e5a40c-f142-55b9-ab15-fc77c0064ce0"
docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
```
Owner: `flex-auth`; production write, operator approval required.
The pin runs with
`--caller-binding secrets-engine=system:serviceaccount:secrets-engine:secrets-engine`
and namespace `secrets-engine` holds only `default`. There is nothing to mint a
token for, and `enforce` today would deny every request rather than
authenticate one.
- Create ServiceAccount `secrets-engine` in namespace `secrets-engine`.
- No `Role`, no `RoleBinding`, no `automountServiceAccountToken`. An SA with no
binding grants nothing in the cluster; its only function is to be the name in
the binding above.
- Add it to the overlay rather than applying it loose, following
`deploy/caller-auth-rbac.yaml`.
Gate: `kubectl -n secrets-engine create token secrets-engine --audience=flex-auth
--duration=10m` returns a token whose `sub` equals the bound principal exactly.
## 2. Run the positive and negative tests and return the receipts
```task
id: FLEX-WP-0023-T02
status: done
docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
priority: high
state_hub_task_id: "79a8d82d-777c-5462-b49d-098f4b7a3b9c"
docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
```
Owner: `flex-auth` to run; `glas-harness` to receive.
Procedure and expectations are in the design doc. One positive (correct SA,
correct audience, inside lifetime, expecting allow at `policy_version: v2`) and
four negatives, each isolating one property:
| # | Credential | Expected under `enforce` |
| --- | --- | --- |
| N1 | no `Authorization` header | 401 |
| N2 | token for `secrets-engine:default` | 403, principal cannot represent system |
| N3 | token minted without `--audience=flex-auth` | 401 |
| N4 | token past `exp` | 401 |
N2 and N4 are the load-bearing ones. N2 proves the binding is a binding rather
than mere possession of a valid cluster token — thousands of ServiceAccounts can
produce a well-formed one and only one may represent `secrets-engine`. N4 proves
the lifetime is enforced by the issuer rather than asserted by the caller.
**Blocked on credential issuance, not on design.** `kubectl create token` is
credential minting and was refused in the 2026-09-06 session that wrote this
plan, correctly. The commands are exact and the expectations are read out of
`internal/callerauth/auth.go`; running them is an operator action. Under `warn`
all four negatives are *allowed*, so running them before `T03` records the
finding rather than a passing test.
Gate: five receipts returned to `glas-harness`, each naming which property it
isolates. Nothing is reported as verified that was not run.
## 3. Flip `callerAuth.mode` to enforce
```task
id: FLEX-WP-0023-T03
status: done
docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
priority: high
state_hub_task_id: "8117c9d8-6efa-5ccf-8ed4-4da2519c3de3"
docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
```
Owner: `flex-auth`; deployment approval required.
- Flip only the `flex-auth-secrets-engine` pin. Do not roll the other three;
they are pinned to their own digests so one change does not move another
consumer (`FLEX-WP-0016`).
- Confirm the warn logs are clean for the adopted identity first, as
`FLEX-WP-0016` did for `ops-warden` — but keep the window short, per the
finding above.
- Re-run `T02`'s four negatives after the flip. Before it they document the
gap; after it they are the test.
Gate: N1N4 refused, positive still allowed, other three pins unchanged.
`glas-harness`'s standing instruction — do not turn warn into enforce before
caller adoption is proved — is satisfied by `T02`, not bypassed by this task.
## 4. Record the authenticated caller in the decision record
```task
id: FLEX-WP-0023-T04
status: todo
priority: high
state_hub_task_id: "c0e4f31a-cc42-5bd2-b938-d140ecd52e1a"
docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
```
Owner: `flex-auth`.
`FLEX-DEC-2026-009`: the verification in `T02`/`T03` is real and leaves no
trace. `flex-auth.decision-record.v1` has no caller field, so a record proves
the subject was allowed and cannot prove the caller who obtained it was
authenticated.
- Add `provenance.caller``mode` (required), `principal`, `audience`,
`not_after`. Not `binding`: the caller is deliberately not decision material,
and putting it there would change `request_digest` and break every consumer's
replay join.
- `mode` is load-bearing. A `principal` recorded under `warn` was observed, not
enforced, and a reader who cannot tell those apart reads an unverified string
as verification. Under `disabled`, emit `{"mode": "disabled"}` — absence
stated, following the `pdp_digest` precedent.
- Capture the reviewed token's `exp` for `not_after`.
`internal/callerauth.Identity` carries only `Username` and `Audiences`, so the
expiry is validated by `TokenReview` and then discarded. This is the one real
code change.
- Update `schemas/decision_envelope.schema.json` and
`docs/decision-record-contract.md`; state in both that `request_digest` is
unaffected, so nobody re-pins in response.
Gate: a decision obtained under `enforce` names its caller and lifetime; the
same request's `request_digest` is byte-identical to the pre-change value.
## 5. Report the gap to gate-house as a v0.8 finding
```task
id: FLEX-WP-0023-T05
docs: v0.8 assent review — four findings, one fail-open gate-house circulated security-layer-model v0.8 for assent. Assent given with four findings; F1 is fail-open and blocking for calling the section 6.4 PEP obligations complete, though not for anyone's adoption — the condition already exists under v0.7 and is not created by this version. F1. "A decision from access-engine" is not a checkable predicate. 6.4 obligation 1 requires a PEP to hold one; obligation 2 supplies a test it is emphatic is mechanical and not a judgement call. That test establishes which request a decision is for and nothing about who issued it, because every input to every digest a PEP can check is either sent by the caller or published. A responder knowing the package id and version reproduces all three exactly. What makes it sharp is internal: 9.4 requires the approval object to carry authenticated entries, and 6.4 obligation 5 then requires each artifact be validated against its owning layer, naming the approval-claim and DecisionEnvelope pair as the live instance. So the obligation is unsatisfiable for half the pair it was written about — a PEP can validate the approval's authenticity and cannot validate the decision's. The asymmetry is argued nowhere. Section 16 already carries this observation one layer up, about the standard's own publication path having no digest or freeze discipline. Applied to the artifact the standard regulates, it is F1. The gap was visible from inside and got recorded against the wrong artifact. F2 answers the absent-vs-unknown question gate-house flagged as ours. Runtime behaviour must be identical — absent fails closed too, and for a stronger reason, since it is the branch reached by discovering your model of your own scope axis was wrong. But omitting the distinction makes totality unfalsifiable: a map with an unknown catch-all satisfies 6.4 obligation 3 vacuously, and its drift test passes by exercising the catch-all rather than the axis. Same defect we published as FLEX-DEC-2026-008 — coverage counted rather than executed, in a stance map instead of a fixture suite. F3. "The published map MUST equal the shipped behaviour, and that equality SHOULD be asserted by a test", in a paragraph arguing drift is worse than no publication. The strongest obligation in 6.4 has the weakest verification, and 13.1 shows four of five rows already test it. F4. 13.1 marks ops-warden's unknown cell non-conformant and states ops-mason's map as simply not published, which is a plainer violation of the same obligation, unmarked in the same table. Also answered their direct question: no, they did not over-read our divergence report. Added that the ruling costs us nothing, since we are not PEP-shaped, so our concurrence is weak evidence and ops-warden's assent is worth more here. FLEX-WP-0023-T05 and FLEX-WP-0024-T04 close. Notes T01-T03 landed from Glas meanwhile with real receipts including an actually-expired token, and that enforce authenticates the caller while leaving F1 untouched — as their own note says. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-07 08:45:39 +02:00
status: done
docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
priority: medium
state_hub_task_id: "d685abfc-cf86-50c8-ad5e-2c04e1531ddd"
docs: operator caller access path, and the caller is absent from the record glas-harness asked for a supported owner access path with authenticated caller binding and lifetime, ruling out Service DNS and a permanent operator token. Both refusals are correct and the answer needs no new mechanism: a TokenRequest token is short-lived, audience-scoped, and bound by exact sub to the ServiceAccount the deployed pin already names. Two findings came out of designing it. First, for an operator caller the pin's NetworkPolicy is not a partial control, it is silent: kubectl port-forward is proxied to the pod's own loopback and no policy selector is consulted. So under warn the operator path is unauthenticated and unfiltered, and the only reason it is not reachable is that nobody has forwarded the port. warn was safe for ops-warden because a workload pin still admitted one pod; here the warn window protects nothing while it runs. enforce is the deliverable. Second, the ServiceAccount named by the deployed binding does not exist. Namespace secrets-engine holds only default, so enforce today would deny every request rather than authenticate one. Third, and this is the contract one: the decision record has no caller field. provenance carries evaluator, mode, policy and registry digests and decision time; binding carries the normalized request. So the four negative tests can all pass and no artifact retains that they passed for the request that mattered. Same seam as FLEX-DEC-2026-008 one layer up — there a tenant was carried into the digest and never compared, visible but not enforced; here a caller is authenticated and never recorded, enforced but not visible. provenance.caller, not binding.caller: the same request from a different authenticated caller must decide identically, so the caller is not replay identity and must not move request_digest. FLEX-DEC-2026-009. Live receipts are not included. kubectl create token is credential minting and was refused in this session, correctly; the commands are exact and the expectations read out of internal/callerauth/auth.go, but nothing is claimed as verified that was not run. Also records the operator's tenant:platform decision (5ed3fb35-eca9-413a-82b9-95171ba85bf6) and verifies that v2 already enforces exactly it — exact string equality, no alias, no normalisation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-06 22:33:21 +02:00
```
Owner: `flex-auth`.
§17 makes the decision-record schema flex-auth's, so this gap is ours to have
missed — and the standard does not ask for what §9.7.2's own argument implies.
§9.7.2 promoted registry-snapshot provenance to a conformance prerequisite
because a decision turning on registry content must be replayable from its own
record. **A decision gated by caller authentication is not auditable from its
own record by the same argument.** Carry it into the outstanding v0.8 assent
review rather than as a separate message.
## Operator execution from Glas — 2026-09-06
Standing production authorization applied to this bounded owner procedure.
Added deploy/secrets-engine-operator-caller.yaml and applied it after server
validation: SA secrets-engine/secrets-engine, automount disabled, no role or
role binding added. Ten-minute TokenRequest token has exact bound sub and
flex-auth audience; token values remain only in proof-process memory.
Positive adoption passed under warn with zero authentication warnings. Helm
lint/server dry-run passed; dedicated release upgraded to revision 3 with
callerAuth.mode=enforce. Correct caller gets 200 allow v2; missing token 401,
wrong principal 403, wrong audience 401. Expiry proof passed against an actually issued token: 401 after its exp
plus 65 seconds, followed by a fresh-token 200 allow v2. Other three Deployment
specs were compared and are unchanged. The temporary forward binds 127.0.0.1
and is removed when the proof finishes. Value-free receipts are in glas-harness:
docs/evidence/GLAS-WP-0015-caller-auth-2026-09-06.json.
The initial proof reader assumed JSON error responses; the server correctly
returns plain-text authentication failures. Corrected the temporary reader and
reran the checks; this was a proof-harness failure, not a production code change.
T04 caller provenance and FLEX-WP-0024 signatures remain separate open work.
An authenticated API-server port-forward authenticates the responder for this
operator path; it does not produce a signed portable decision artifact.
Final receipt: T01T03 complete; N4 expired-token 401 at epoch 1788730498,
fresh-token positive 200 at 1788730499. Temporary forward closed and proof
process exited, discarding its in-memory credentials. No port-forward remains
as an implicit runtime dependency. T04 and T05 remain open; plan stays active.
docs: v0.8 assent review — four findings, one fail-open gate-house circulated security-layer-model v0.8 for assent. Assent given with four findings; F1 is fail-open and blocking for calling the section 6.4 PEP obligations complete, though not for anyone's adoption — the condition already exists under v0.7 and is not created by this version. F1. "A decision from access-engine" is not a checkable predicate. 6.4 obligation 1 requires a PEP to hold one; obligation 2 supplies a test it is emphatic is mechanical and not a judgement call. That test establishes which request a decision is for and nothing about who issued it, because every input to every digest a PEP can check is either sent by the caller or published. A responder knowing the package id and version reproduces all three exactly. What makes it sharp is internal: 9.4 requires the approval object to carry authenticated entries, and 6.4 obligation 5 then requires each artifact be validated against its owning layer, naming the approval-claim and DecisionEnvelope pair as the live instance. So the obligation is unsatisfiable for half the pair it was written about — a PEP can validate the approval's authenticity and cannot validate the decision's. The asymmetry is argued nowhere. Section 16 already carries this observation one layer up, about the standard's own publication path having no digest or freeze discipline. Applied to the artifact the standard regulates, it is F1. The gap was visible from inside and got recorded against the wrong artifact. F2 answers the absent-vs-unknown question gate-house flagged as ours. Runtime behaviour must be identical — absent fails closed too, and for a stronger reason, since it is the branch reached by discovering your model of your own scope axis was wrong. But omitting the distinction makes totality unfalsifiable: a map with an unknown catch-all satisfies 6.4 obligation 3 vacuously, and its drift test passes by exercising the catch-all rather than the axis. Same defect we published as FLEX-DEC-2026-008 — coverage counted rather than executed, in a stance map instead of a fixture suite. F3. "The published map MUST equal the shipped behaviour, and that equality SHOULD be asserted by a test", in a paragraph arguing drift is worse than no publication. The strongest obligation in 6.4 has the weakest verification, and 13.1 shows four of five rows already test it. F4. 13.1 marks ops-warden's unknown cell non-conformant and states ops-mason's map as simply not published, which is a plainer violation of the same obligation, unmarked in the same table. Also answered their direct question: no, they did not over-read our divergence report. Added that the ruling costs us nothing, since we are not PEP-shaped, so our concurrence is weak evidence and ops-warden's assent is worth more here. FLEX-WP-0023-T05 and FLEX-WP-0024-T04 close. Notes T01-T03 landed from Glas meanwhile with real receipts including an actually-expired token, and that enforce authenticates the caller while leaving F1 untouched — as their own note says. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014aQMM1dPXaPiXVn6DwwtLd Assistant: claude-code Assistant-Model: opus Assistant-Process: 715613@bnt-lap001 Assistant-Session: fabd95c1-4c9e-4080-8849-8707ae025f80
2026-09-07 08:45:39 +02:00
## T05 closed — 2026-09-07
**Done.** Carried into the v0.8 assent round as `FLEX-DEC-2026-011`
finding F1, and it sharpened on the way: the standard is not merely silent on
responder authenticity. §6.4 obligation 1 uses "a decision from `access-engine`"
as though it named a verifiable property, and obligation 2 supplies a mechanical
test that a careful implementer reads as discharging it. §9.4 meanwhile requires
the approval object to carry authenticated entries, so obligation 5 — validate
each artifact against its owning layer — is unsatisfiable for half the pair it
was written about. The caller-provenance gap folded into the same finding.
Glas's T01T03 execution does not weaken F1 and its own note says so: an
authenticated API-server port-forward authenticates the responder *for that
operator path* and produces no signed portable artifact. `enforce` authenticates
the caller to the PDP. Nothing yet authenticates the PDP to the caller off that
one channel, which is what `FLEX-WP-0024` is for.