informed-decision and approval-engine both asked to hear problems with the human
approver registration now rather than at handover. Checking their requested shape
against the source rather than agreeing it on paper turned up three things.
Scope gap, accepted: [openid, approval:approve] cannot render a decision, since
GET /v1/approvals/{id} and /claim both need approval:read -- the surface could
submit an entry it was never able to display. Published
[openid, approval:read, approval:approve]. Reading through a service identity was
the alternative and is worse: it weakens the evidence-of-what-this-person-saw
claim the component exists to make. approval:consume stays excluded.
Assurance shape, published and a defect fixed. Both asked for a documented shape
and KeyCape already emitted one, so it is written down rather than renegotiated.
Writing it down surfaced that `at` was the token mint time rather than the
authentication time. Those differ by hours whenever a browser session is reused,
and approval-engine persists this object verbatim as the only downstream record
that MFA happened -- so a stored approval could have evidenced MFA at a moment
the person proved nothing. PKCESession.AuthTime now carries the original login
instant through session reuse, with mint time as the fallback.
Blocker found before anyone built on it: a human token cannot carry
tenant:platform. effectiveTenant resolves the human tenant from the directory
user, no adapter populates User.Tenant, and the per-client tenant field is read
only on the client_credentials path -- so every human token defaults to
tenant:coulomb, which approval-engine refuses by exact string equality. It would
have presented as a failed approval rather than a registration defect. Two
resolutions sent to the owners and neither implemented here: the choice decides
whether a human's tenant is a property of the person or of the registration, and
that is not KeyCape's alone to make.
Also recorded ops-warden's answers to KEY-WP-0014-T04, including their finding
that `warden plan` returns `autonomous` for a need containing generate and
CAS-write, because it has no read-versus-mutate intent. Their standing
instruction -- treat a warden plan verdict on any write, rotate or provision need
as unreliable until WARDEN-WP-0038 lands -- is recorded in the workplan rather
than left in an inbox.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016uV8zoCKpA1WRAxsKRYbdH
Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 1182213@bnt-lap001
Assistant-Session: 966597b9-ae61-46a4-8b9e-1594ab3ec4ad
KEY-WP-0013-T02 and KEY-WP-0014-T04 stay blocked on custody and on ops-warden,
but each contains a KeyCape-owned piece that had been left as prose. T02 requires
proving "live JWKS verification and denied excess scopes without logging values";
T04 step 4 requires verifying a rotated secret, refusing its predecessor and
refusing excess scope. Both were describable and neither was runnable, so the
proof would have been improvised by hand against production at the moment custody
lands -- the worst possible time for it.
keycape verify-client does it in one command. Per registration it checks
discovery origin, the client_credentials exchange and its RS256 signature against
the deployed JWKS, exact sub/tenant/roles/principal_type, that every -deny-scope
is refused, and that the token carries no scope that was not requested. That last
check is a real gap: the caller commands prove every requested scope was granted,
never that nothing extra came back. -previous-secret-env additionally requires
the predecessor to be refused, and treats an unchanged secret as a rotation that
did not happen.
Nothing is written to disk and no value is printed. Failures name the claim, not
the observed value, so running this against production cannot turn a verification
into a disclosure. Every check runs before it reports, so one failure does not
hide the rest.
The exact invocation for each approval client is recorded in the verification
block of the provisioning packet, so custody admission hands back a command
rather than a description.
Tests cover the passing case, an over-broad registration, a live predecessor, an
identical rotation and output non-disclosure; neutering mustFail makes the suite
fail, so the checks have teeth.
Also recorded in KEY-WP-0014: read from ops-warden's catalog rather than waiting
for a reply, key-cape-oidc-login is asked-and-waiting on us since 2026-08-28 and
is a pointer lane with no programmatic consumers, and the rapp-qonto-keycape-client
blocker citing an absent native exchange command went stale when service-token
shipped on 2026-09-05.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016uV8zoCKpA1WRAxsKRYbdH
Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 1182213@bnt-lap001
Assistant-Session: 966597b9-ae61-46a4-8b9e-1594ab3ec4ad
G10 waits on custody and platform owners and cannot close from here. What was
doable: verify the handoffs actually went out, replace a remembered live state
with an observed one, and find out whether main is safe to deploy. The last
question found a defect in this repository's own recent work.
Handoffs verified independently rather than trusted: all seven messages are in
the hub with receipt ids. This gap was reopened once for claimed-but-unsent
delivery, so the claim deserved the same scrutiny.
Live state read from the cluster read-only: image main-153258b, only the Qonto
secret materialized so the approval clients remain unprovisioned, four registered
clients, no tenantEngine block. That also corrects an earlier claim of mine --
the deployed config sets userOU explicitly, so the KEY-WP-0023 default fix was
never a production issue.
The precondition: KEY-WP-0019 discovers the expected issuer from
authelia.tokenBaseURL, and the deployed Authelia derives its advertised issuer
from the request Host, advertising the in-cluster address to KeyCape and the
browser-facing one to browsers. Verification fails closed, so a mismatch breaks
every human login and looks like a broken login rather than a misconfiguration.
Which value the token carries needs a real login against production to settle and
was not determined here.
Two mitigations: docs/operations.md documents pinning authelia.issuer and
jwksUrl, with the curl that reveals what the provider advertises for a given
Host; and the authentication failure event now carries a specific reason, so
id_token_issuer_mismatch is distinguishable from a signature failure or an
unreachable key set. The browser still learns nothing.
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
Closes gap G08. /healthz returned a constant without probing anything, the server
called ListenAndServe with no signal handling, and the operational limits of
in-memory state, startup-loaded keys and local-only logout lived in code comments
rather than anywhere an operator would look.
/readyz probes LLDAP, Authelia and privacyIDEA; /healthz stays liveness and
probes nothing. Keeping them distinct matters: wiring liveness to dependency
health means an orchestrator restarts KeyCape when a dependency blinks, and a
restart also discards every in-flight login, so the reaction is worse than the
condition it reacts to.
LLDAP is probed with a bind rather than a dial, since a rotated or revoked
service password leaves the port open and every lookup failing -- exactly what
readiness should catch and exactly what a dial would miss. The response names the
failing check but never the reason: the endpoint is unauthenticated and upstream
error text carries hostnames and sometimes credentials-in-URLs. Results are
cached for 2s so an unauthenticated endpoint cannot be used to drive unbounded
upstream traffic, and probes run concurrently under a 3s bound so a hung
dependency makes the endpoint answer rather than hang with it.
SIGTERM and SIGINT now drain in-flight requests for 15s, under the 30s read/write
timeouts so a stuck request cannot outlive the window before SIGKILL.
docs/operations.md states the single-replica topology and why, and three limits
easy to get wrong: the constant key-1 key ID makes same-kid rotation a trap for
consumers caching JWKS, removing a client does not revoke its issued tokens, and
/logout is local only. No throughput figures are given, since nothing here
benchmarks KeyCape. Shared storage and refresh tokens stay excluded, as G08
allows.
Verified in the running executable: 503 naming all three checks failed while
/healthz returned 200, the LLDAP check flipping to ok once started, and 40/40
requests succeeding across a SIGTERM.
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
KEY-WP-0016 changed /token and /userinfo behaviour with no consumer-facing note;
nothing in docs/ mentioned redirect_uri, so the change would have reached a
deployment silently.
States what an exchange must now send, who is affected and how to roll out.
Every browser registration in dev-config is public with an authorization_code
grant, so only the redirect_uri requirement can affect them; the realistic
failure is a client that sends it to /authorize and omits it at /token, which
has not been observed against a live consumer.
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
Operator decision 5ed3fb35-eca9-413a-82b9-95171ba85bf6 accepts tenant:platform
as the platform management tenant for the Glas approval chain, requiring exact
spelling across the approval store, the service-client JWT claim and the
lifecycle CheckRequest.
Changes the tenant field on secrets-engine-approval and approval-engine-operator
only, in the registration fixture and the provisioning packet. Unrelated clients
and the human directory default keep tenant:coulomb, and no audience, scope,
subject, role, lifetime or MFA grant changes.
Adds issuance evidence that the approval shape emits tenant:platform exactly and
never an alias the caller requests, that the OpenBao client gains no
cross-tenant reach, and a fixture guard pinning every reviewed client's tenant.
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
Answers the GLAS-WP-0015 tenant-alignment request. The JWT tenant is
tenant:coulomb for all four reviewed service registrations; the approval store
tenant and the policy tenant belong to other owners and KeyCape does no
normalization between them. No mapping is invented and no live registration
changes.
Adds regression tests proving the tenant claim is bound at registration and is
not influenced by request parameters, that distinct registrations never carry
each other's tenant, and that human tokens fall back to the platform tenant
rather than an empty claim.
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
