fix(decision): registry facts win over caller-supplied attributes
secrets-engine's first live request rejected our allow: binding.
request_digest is computed over material they never sent, because we
enrich subject and resource from the registry before hashing. Answering
that meant reading the enrichment path, which had a worse defect in it.
Enrichment was additive-if-absent — addAttribute wrote a registry value
only where the request had no value for that key. So where a caller
supplied a key, the caller's value won and the registry's never applied.
Every registry ceiling and allowlist was advisory. Verified against the
shipped ops-warden package, each one added key on an otherwise-denied
request:
max_ttl_hours: 99 registry says 8 -> allowed a 12h certificate
allowed_principals registry allowlist -> disallowed_principal bypassed
allowed_subjects registry allowlist -> unknown_subject bypassed
The third is the one to read twice: a subject the registry does not know
authorized itself by naming itself in the allowlist it was being checked
against.
Not remotely reachable today — the PEP builds the CheckRequest,
ops-warden sends no resource.attributes, and enforce admits one identity.
It is a defence-in-depth failure: any path that lets attacker-influenced
data into a CheckRequest field became a full policy bypass rather than a
bounded input problem. Callers sending resource.attributes is not
hypothetical; secrets-engine does it on every request.
Registry facts now win, and diagnostics.registry_overrode names every
displaced key, because a registry that silently discards a contradicting
claim hides that a caller asserted authority it did not have.
subject.type is carved out, and the reason is a finding of its own.
Making the registry win there denied every secrets-engine allow: the
registry's type is CARING vocabulary (Human, Agent, Automation, Service)
and the request's is the protected system's actor vocabulary (service,
adm, agt, atm). Two fields sharing a name; substituting one for the other
is translation rather than identity, which GH-DEC-2026-008 ruled against.
Note what surfaced it — the registry's type had been dead data since the
field existed, because the caller's value always won.
Also publishes binding.submitted_request_digest, over the request exactly
as sent. request_digest was published as the consumer replay test and
cannot be one. Nothing is lost hashing the pre-enrichment form:
enrichment is a function of the request and the snapshot, and
registry_snapshot_digest already pins the snapshot.
Existing pins do not move. All three replay fixtures' request_digest and
approval_binding_digest values are byte-identical — those requests
contradict no registry fact. A field to add, not a value to correct.
Regression tests verified failing against the old behaviour before being
kept. FLEX-DEC-2026-012; FLEX-WP-0025 carries the residual, that a policy
still cannot tell a fact from an assertion.
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 13:43:33 +02:00
|
|
|
|
---
|
|
|
|
|
|
id: FLEX-WP-0025
|
|
|
|
|
|
type: workplan
|
|
|
|
|
|
title: "A policy cannot tell a registry fact from a caller assertion"
|
|
|
|
|
|
domain: infotech
|
|
|
|
|
|
repo: flex-auth
|
|
|
|
|
|
status: ready
|
|
|
|
|
|
owner: claude
|
|
|
|
|
|
topic_slug: netkingdom
|
|
|
|
|
|
planning_priority: P1
|
|
|
|
|
|
planning_order: 250
|
|
|
|
|
|
depends_on_workplans:
|
|
|
|
|
|
- FLEX-WP-0021
|
|
|
|
|
|
related_workplans:
|
|
|
|
|
|
- FLEX-WP-0022
|
|
|
|
|
|
- FLEX-WP-0024
|
|
|
|
|
|
created: "2026-09-07"
|
|
|
|
|
|
updated: "2026-09-07"
|
2026-09-07 13:46:31 +02:00
|
|
|
|
state_hub_workstream_id: "f9a657ce-67b4-5d25-9933-e0fcb2c20b1c"
|
fix(decision): registry facts win over caller-supplied attributes
secrets-engine's first live request rejected our allow: binding.
request_digest is computed over material they never sent, because we
enrich subject and resource from the registry before hashing. Answering
that meant reading the enrichment path, which had a worse defect in it.
Enrichment was additive-if-absent — addAttribute wrote a registry value
only where the request had no value for that key. So where a caller
supplied a key, the caller's value won and the registry's never applied.
Every registry ceiling and allowlist was advisory. Verified against the
shipped ops-warden package, each one added key on an otherwise-denied
request:
max_ttl_hours: 99 registry says 8 -> allowed a 12h certificate
allowed_principals registry allowlist -> disallowed_principal bypassed
allowed_subjects registry allowlist -> unknown_subject bypassed
The third is the one to read twice: a subject the registry does not know
authorized itself by naming itself in the allowlist it was being checked
against.
Not remotely reachable today — the PEP builds the CheckRequest,
ops-warden sends no resource.attributes, and enforce admits one identity.
It is a defence-in-depth failure: any path that lets attacker-influenced
data into a CheckRequest field became a full policy bypass rather than a
bounded input problem. Callers sending resource.attributes is not
hypothetical; secrets-engine does it on every request.
Registry facts now win, and diagnostics.registry_overrode names every
displaced key, because a registry that silently discards a contradicting
claim hides that a caller asserted authority it did not have.
subject.type is carved out, and the reason is a finding of its own.
Making the registry win there denied every secrets-engine allow: the
registry's type is CARING vocabulary (Human, Agent, Automation, Service)
and the request's is the protected system's actor vocabulary (service,
adm, agt, atm). Two fields sharing a name; substituting one for the other
is translation rather than identity, which GH-DEC-2026-008 ruled against.
Note what surfaced it — the registry's type had been dead data since the
field existed, because the caller's value always won.
Also publishes binding.submitted_request_digest, over the request exactly
as sent. request_digest was published as the consumer replay test and
cannot be one. Nothing is lost hashing the pre-enrichment form:
enrichment is a function of the request and the snapshot, and
registry_snapshot_digest already pins the snapshot.
Existing pins do not move. All three replay fixtures' request_digest and
approval_binding_digest values are byte-identical — those requests
contradict no registry fact. A field to add, not a value to correct.
Regression tests verified failing against the old behaviour before being
kept. FLEX-DEC-2026-012; FLEX-WP-0025 carries the residual, that a policy
still cannot tell a fact from an assertion.
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 13:43:33 +02:00
|
|
|
|
---
|
|
|
|
|
|
|
|
|
|
|
|
# FLEX-WP-0025 — A policy cannot tell a registry fact from a caller assertion
|
|
|
|
|
|
|
|
|
|
|
|
Opened by `FLEX-DEC-2026-012`. That decision closed three verified privilege
|
|
|
|
|
|
escalations by making registry facts win over caller-supplied attributes. It did
|
|
|
|
|
|
not close the shape that produced them.
|
|
|
|
|
|
|
|
|
|
|
|
`input.resource.attributes.max_ttl_hours` reaches a Rego rule as one value in one
|
|
|
|
|
|
map. **Nothing in that input says whether the registry asserted it or the caller
|
|
|
|
|
|
did.** Precedence decides which value survives; it does not let a policy author
|
|
|
|
|
|
require that a ceiling came from the registry at all.
|
|
|
|
|
|
|
|
|
|
|
|
## The two residuals precedence did not close
|
|
|
|
|
|
|
|
|
|
|
|
**1. A key the manifest omits is caller input.** Registry facts win only where
|
|
|
|
|
|
the registry has a value. A package branching on an attribute its manifest does
|
|
|
|
|
|
not declare is reading the caller, and a manifest that stops declaring a key
|
|
|
|
|
|
silently hands that key back. Registering the resource is not sufficient — the
|
|
|
|
|
|
specific key must be present.
|
|
|
|
|
|
|
|
|
|
|
|
**2. `subject.type` is two fields sharing a name.** The registry's `type` is
|
|
|
|
|
|
CARING vocabulary (`Human`, `Agent`, `Automation`, `Service`); the request's is
|
|
|
|
|
|
the protected system's actor vocabulary (`service`, `adm`, `agt`, `atm`).
|
|
|
|
|
|
Policies compare against the latter. Making the registry win here denied every
|
|
|
|
|
|
`secrets-engine` allow, because `"Service"` is not `"service"`, so the exception
|
|
|
|
|
|
stands — and a caller can still assert its own `subject.type` while packages
|
|
|
|
|
|
branch on it.
|
|
|
|
|
|
|
|
|
|
|
|
Note how the second surfaced: the registry's `type` had been dead data since the
|
|
|
|
|
|
field existed, because the caller's value always won. **A defect is invisible
|
|
|
|
|
|
while the value it produces is never read.**
|
|
|
|
|
|
|
|
|
|
|
|
## 1. Decide the shape
|
|
|
|
|
|
|
|
|
|
|
|
```task
|
|
|
|
|
|
id: FLEX-WP-0025-T01
|
|
|
|
|
|
status: todo
|
|
|
|
|
|
priority: high
|
2026-09-07 13:46:31 +02:00
|
|
|
|
state_hub_task_id: "05c6a85d-ee3d-5875-bea4-e9a9c5382e2e"
|
fix(decision): registry facts win over caller-supplied attributes
secrets-engine's first live request rejected our allow: binding.
request_digest is computed over material they never sent, because we
enrich subject and resource from the registry before hashing. Answering
that meant reading the enrichment path, which had a worse defect in it.
Enrichment was additive-if-absent — addAttribute wrote a registry value
only where the request had no value for that key. So where a caller
supplied a key, the caller's value won and the registry's never applied.
Every registry ceiling and allowlist was advisory. Verified against the
shipped ops-warden package, each one added key on an otherwise-denied
request:
max_ttl_hours: 99 registry says 8 -> allowed a 12h certificate
allowed_principals registry allowlist -> disallowed_principal bypassed
allowed_subjects registry allowlist -> unknown_subject bypassed
The third is the one to read twice: a subject the registry does not know
authorized itself by naming itself in the allowlist it was being checked
against.
Not remotely reachable today — the PEP builds the CheckRequest,
ops-warden sends no resource.attributes, and enforce admits one identity.
It is a defence-in-depth failure: any path that lets attacker-influenced
data into a CheckRequest field became a full policy bypass rather than a
bounded input problem. Callers sending resource.attributes is not
hypothetical; secrets-engine does it on every request.
Registry facts now win, and diagnostics.registry_overrode names every
displaced key, because a registry that silently discards a contradicting
claim hides that a caller asserted authority it did not have.
subject.type is carved out, and the reason is a finding of its own.
Making the registry win there denied every secrets-engine allow: the
registry's type is CARING vocabulary (Human, Agent, Automation, Service)
and the request's is the protected system's actor vocabulary (service,
adm, agt, atm). Two fields sharing a name; substituting one for the other
is translation rather than identity, which GH-DEC-2026-008 ruled against.
Note what surfaced it — the registry's type had been dead data since the
field existed, because the caller's value always won.
Also publishes binding.submitted_request_digest, over the request exactly
as sent. request_digest was published as the consumer replay test and
cannot be one. Nothing is lost hashing the pre-enrichment form:
enrichment is a function of the request and the snapshot, and
registry_snapshot_digest already pins the snapshot.
Existing pins do not move. All three replay fixtures' request_digest and
approval_binding_digest values are byte-identical — those requests
contradict no registry fact. A field to add, not a value to correct.
Regression tests verified failing against the old behaviour before being
kept. FLEX-DEC-2026-012; FLEX-WP-0025 carries the residual, that a policy
still cannot tell a fact from an assertion.
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 13:43:33 +02:00
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
|
|
Owner: `flex-auth`.
|
|
|
|
|
|
|
|
|
|
|
|
The obvious move is separate namespaces — `input.resource.registry.*` for facts
|
|
|
|
|
|
and `input.resource.asserted.*` for caller input, with `attributes` retained as
|
|
|
|
|
|
the merged view. That breaks no package on day one and lets a rule that enforces
|
|
|
|
|
|
a ceiling opt into the fact side.
|
|
|
|
|
|
|
|
|
|
|
|
Weigh against it: a third view invites a package to read the merged one by
|
|
|
|
|
|
habit, which is the status quo with extra steps. A stricter alternative is to
|
|
|
|
|
|
stop merging at all and require packages to name a side, accepting a breaking
|
|
|
|
|
|
change across six published packages.
|
|
|
|
|
|
|
|
|
|
|
|
Do not decide this from the shape alone. Read what the six packages actually
|
|
|
|
|
|
branch on first, and write down which of their reads are ceilings or allowlists
|
|
|
|
|
|
(must be facts), which are genuinely caller-proposed (`ttl_hours`, `purpose`),
|
|
|
|
|
|
and which are ambiguous. The ambiguous set is the real design input.
|
|
|
|
|
|
|
|
|
|
|
|
Gate: the chosen shape is written down with the vocabulary collision addressed,
|
|
|
|
|
|
before any package changes.
|
|
|
|
|
|
|
|
|
|
|
|
## 2. Audit every package for ceilings read from undeclared keys
|
|
|
|
|
|
|
|
|
|
|
|
```task
|
|
|
|
|
|
id: FLEX-WP-0025-T02
|
|
|
|
|
|
status: todo
|
|
|
|
|
|
priority: high
|
2026-09-07 13:46:31 +02:00
|
|
|
|
state_hub_task_id: "fc61c8a7-347b-59d2-b6a6-a72c5c1cdb08"
|
fix(decision): registry facts win over caller-supplied attributes
secrets-engine's first live request rejected our allow: binding.
request_digest is computed over material they never sent, because we
enrich subject and resource from the registry before hashing. Answering
that meant reading the enrichment path, which had a worse defect in it.
Enrichment was additive-if-absent — addAttribute wrote a registry value
only where the request had no value for that key. So where a caller
supplied a key, the caller's value won and the registry's never applied.
Every registry ceiling and allowlist was advisory. Verified against the
shipped ops-warden package, each one added key on an otherwise-denied
request:
max_ttl_hours: 99 registry says 8 -> allowed a 12h certificate
allowed_principals registry allowlist -> disallowed_principal bypassed
allowed_subjects registry allowlist -> unknown_subject bypassed
The third is the one to read twice: a subject the registry does not know
authorized itself by naming itself in the allowlist it was being checked
against.
Not remotely reachable today — the PEP builds the CheckRequest,
ops-warden sends no resource.attributes, and enforce admits one identity.
It is a defence-in-depth failure: any path that lets attacker-influenced
data into a CheckRequest field became a full policy bypass rather than a
bounded input problem. Callers sending resource.attributes is not
hypothetical; secrets-engine does it on every request.
Registry facts now win, and diagnostics.registry_overrode names every
displaced key, because a registry that silently discards a contradicting
claim hides that a caller asserted authority it did not have.
subject.type is carved out, and the reason is a finding of its own.
Making the registry win there denied every secrets-engine allow: the
registry's type is CARING vocabulary (Human, Agent, Automation, Service)
and the request's is the protected system's actor vocabulary (service,
adm, agt, atm). Two fields sharing a name; substituting one for the other
is translation rather than identity, which GH-DEC-2026-008 ruled against.
Note what surfaced it — the registry's type had been dead data since the
field existed, because the caller's value always won.
Also publishes binding.submitted_request_digest, over the request exactly
as sent. request_digest was published as the consumer replay test and
cannot be one. Nothing is lost hashing the pre-enrichment form:
enrichment is a function of the request and the snapshot, and
registry_snapshot_digest already pins the snapshot.
Existing pins do not move. All three replay fixtures' request_digest and
approval_binding_digest values are byte-identical — those requests
contradict no registry fact. A field to add, not a value to correct.
Regression tests verified failing against the old behaviour before being
kept. FLEX-DEC-2026-012; FLEX-WP-0025 carries the residual, that a policy
still cannot tell a fact from an assertion.
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 13:43:33 +02:00
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
|
|
Owner: `flex-auth`.
|
|
|
|
|
|
|
|
|
|
|
|
Mechanical and worth doing before T01 lands, because it is the live residual.
|
|
|
|
|
|
For each published package, list every `input.*.attributes.<key>` read, and check
|
|
|
|
|
|
the key is declared in the manifest for every resource that package can be asked
|
|
|
|
|
|
about.
|
|
|
|
|
|
|
|
|
|
|
|
Any ceiling or allowlist read from a key the manifest does not declare is a live
|
|
|
|
|
|
escalation of the `FLEX-DEC-2026-012` shape. Fix by declaring the key, not by
|
|
|
|
|
|
changing the rule.
|
|
|
|
|
|
|
|
|
|
|
|
Gate: a table of package × key × declared-in-manifest, with no ceiling or
|
|
|
|
|
|
allowlist unbacked. Add it to `docs/request-enrichment.md`.
|
|
|
|
|
|
|
|
|
|
|
|
## 3. Make the review obligation enforceable rather than written
|
|
|
|
|
|
|
|
|
|
|
|
```task
|
|
|
|
|
|
id: FLEX-WP-0025-T03
|
|
|
|
|
|
status: wait
|
|
|
|
|
|
priority: medium
|
2026-09-07 13:46:31 +02:00
|
|
|
|
state_hub_task_id: "8ee5baf9-b364-5d3e-9c49-d0c9eb014c10"
|
fix(decision): registry facts win over caller-supplied attributes
secrets-engine's first live request rejected our allow: binding.
request_digest is computed over material they never sent, because we
enrich subject and resource from the registry before hashing. Answering
that meant reading the enrichment path, which had a worse defect in it.
Enrichment was additive-if-absent — addAttribute wrote a registry value
only where the request had no value for that key. So where a caller
supplied a key, the caller's value won and the registry's never applied.
Every registry ceiling and allowlist was advisory. Verified against the
shipped ops-warden package, each one added key on an otherwise-denied
request:
max_ttl_hours: 99 registry says 8 -> allowed a 12h certificate
allowed_principals registry allowlist -> disallowed_principal bypassed
allowed_subjects registry allowlist -> unknown_subject bypassed
The third is the one to read twice: a subject the registry does not know
authorized itself by naming itself in the allowlist it was being checked
against.
Not remotely reachable today — the PEP builds the CheckRequest,
ops-warden sends no resource.attributes, and enforce admits one identity.
It is a defence-in-depth failure: any path that lets attacker-influenced
data into a CheckRequest field became a full policy bypass rather than a
bounded input problem. Callers sending resource.attributes is not
hypothetical; secrets-engine does it on every request.
Registry facts now win, and diagnostics.registry_overrode names every
displaced key, because a registry that silently discards a contradicting
claim hides that a caller asserted authority it did not have.
subject.type is carved out, and the reason is a finding of its own.
Making the registry win there denied every secrets-engine allow: the
registry's type is CARING vocabulary (Human, Agent, Automation, Service)
and the request's is the protected system's actor vocabulary (service,
adm, agt, atm). Two fields sharing a name; substituting one for the other
is translation rather than identity, which GH-DEC-2026-008 ruled against.
Note what surfaced it — the registry's type had been dead data since the
field existed, because the caller's value always won.
Also publishes binding.submitted_request_digest, over the request exactly
as sent. request_digest was published as the consumer replay test and
cannot be one. Nothing is lost hashing the pre-enrichment form:
enrichment is a function of the request and the snapshot, and
registry_snapshot_digest already pins the snapshot.
Existing pins do not move. All three replay fixtures' request_digest and
approval_binding_digest values are byte-identical — those requests
contradict no registry fact. A field to add, not a value to correct.
Regression tests verified failing against the old behaviour before being
kept. FLEX-DEC-2026-012; FLEX-WP-0025 carries the residual, that a policy
still cannot tell a fact from an assertion.
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 13:43:33 +02:00
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
|
|
Owner: `flex-auth`.
|
|
|
|
|
|
|
|
|
|
|
|
`docs/request-enrichment.md` now says a policy enforcing a ceiling MUST read a
|
|
|
|
|
|
key its manifest declares. A review obligation nothing checks is the shape
|
|
|
|
|
|
`FLEX-DEC-2026-008` was about.
|
|
|
|
|
|
|
|
|
|
|
|
Extend `flex-auth validate` to report attribute keys a package reads that its
|
|
|
|
|
|
fixtures' registry never supplies. It cannot prove intent — it does not know
|
|
|
|
|
|
which reads are ceilings — but it can surface every read that resolves only from
|
|
|
|
|
|
caller input, which is the set a reviewer must look at.
|
|
|
|
|
|
|
|
|
|
|
|
Gate: the check runs in `validate` and flags a package whose ceiling key is
|
|
|
|
|
|
undeclared, demonstrated against a deliberately broken fixture package.
|