key-cape/workplans/KEY-WP-0028-browser-client-field-validation.md
tegwick 1dc11b8448 Record live custody admission and re-verify the new rule against it
Read-only inspection shows KEY-WP-0013-T02's custody gate has moved since
yesterday: both approval clients are registered in sso/keycape-config, both
secrets exist in the namespace and are wired into the pod, and the image is
pinned to the digest T06 published. The live issuer passes every conformance
check that needs no credential.

That does not establish the claim contract. Subject, tenant, roles, scopes,
lifetime and excess-scope denial are what keycape verify-client proves, and it
needs the client secret in the environment. Reading either secret is client-side
retrieval, which the provisioning packet records as not admitted, so that run is
an attended operator action. T02 now waits on the run, not on custody.

Also re-verified the KEY-WP-0028 rule against the deployed config, after a peer
noted it fails closed at startup with a rollout imminent. The three browser
clients there carry neither serviceSubject nor roles; the three service clients
carry both legitimately. Kept as errors rather than warnings on that evidence.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NV9oijZukGyGbRQGGKnK4P

Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 713576@bnt-lap001
Assistant-Session: 384c511d-9bce-4cb8-a676-2aef6c0c8df6
2026-09-09 14:44:38 +02:00

79 lines
3.3 KiB
Markdown

---
id: KEY-WP-0028
type: workplan
title: "Reject service-identity fields that a browser client silently ignores"
domain: infotech
repo: key-cape
status: finished
owner: claude
topic_slug: browser-client-field-validation
created: "2026-09-09"
updated: "2026-09-09"
state_hub_workstream_id: "6d0cb4fd-17da-5466-8011-5ade1ac46171"
---
`serviceSubject` and `roles` are read only on the `client_credentials` path. On a
browser client they are accepted at startup and then ignored: the subject and
roles come from the directory user. A registration that looks effective and is
not fails later, somewhere else, as a downstream rejection rather than as a
registration defect.
`tokenLifetime` was already rejected this way for non-service clients, so the
rule exists — it was just incomplete.
## Reject the ignored fields
```task
id: KEY-WP-0028-T01
status: done
priority: medium
state_hub_task_id: "e4bce8c7-0e60-55e8-ad32-a3a0c4e46044"
```
Config validation now rejects `serviceSubject` and `roles` on any client that
does not use `client_credentials`, naming where the value actually comes from.
Nothing in `config/dev-config.yaml`, the example fixture or the live deployment
sets either field on a browser client, so this breaks no existing configuration —
checked against all three rather than assumed.
Re-verified against `sso/keycape-config` on 2026-09-09, after a peer raised that
this validation fails closed at startup and a rollout is imminent, so a stray
field would surface as KeyCape failing to boot. The deployed config now holds six
clients: `demo-app`, `netkingdom-bootstrap-console` and `openbao-admin` are
browser clients carrying neither field, while `rapp-qonto-client`,
`secrets-engine-approval` and `approval-engine-operator` carry both legitimately
on the `client_credentials` path. The rule rejects none of them. Kept as errors
rather than downgraded to warnings on that evidence: a warning for a field that
does nothing is the state this change exists to end.
**`tenant` is deliberately excluded from the rule**, and a test pins that. A
browser client may declare a tenant; `humanTenant` resolves it against the
directory and refuses issuance when the two disagree (KEY-WP-0013-T05). An
earlier version of this change rejected `tenant` too, which would have made that
feature unusable.
## Correction
```task
id: KEY-WP-0028-T02
status: done
priority: medium
state_hub_task_id: "5ab951f8-2f9c-522c-b6ea-7f6d572cd618"
```
This work started from KEY-WP-0013-T05's blocker paragraph, which described a
human token as unable to carry `tenant:platform`. That was true when written and
had already been fixed in a concurrent session by the time this task began; the
paragraph was read as current state rather than as a dated record.
Two consequences, both corrected within the session:
- the `tenant` rejection was removed before it was committed, and a test now
asserts a browser client carrying `tenant: tenant:platform` validates cleanly;
- a hub decision (`0145ab57`) recording the tenant question as open and
unimplemented was withdrawn with a rationale pointing at the implementation.
Nothing was built on it.
The lesson worth keeping: in a repository where several sessions work at once,
blocker prose ages faster than the code it describes. Re-read the source before
acting on a workplan's description of it.