ops-warden/layer.yaml
tegwick 94f32bd160
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 1s
Review layer model v0.6; publish the PEP stance map §6.4 requires
All three v0.4 findings were acted on — §9.1 split into pending/declared-gap and
§5's scope rule adopted as recommended and credited, and §9.6 ruled via the
load-bearing/attributive distinction with ops-warden's `# audit must not block
signing` named as the estate's live example.

Checked the favourable ruling rather than accepting it. §9.6's test is "no
control branches on its presence": the only consumer of audit.jsonl is `warden
activity`, which displays. Nothing gates on a signing record, so the lane is
genuinely attributive. AuditTrail.md now records the ruling instead of the open
question, and states that the trade must be revisited if a control ever gates on
the trail.

CONFORMANCE ACTION. §6.4 obligation 3 requires a stance map "published rather
than held in code", and requires every PEP-shaped consumer to publish one so the
maps can be inventoried — naming ADR-0009 as the reference shape. ops-warden was
not doing it: the map lived in PolicyConfig.failure_modes, a dataclass default.
Not a code comment, but not published either.

pep-stance.yaml publishes it, and the test asserts the published map EQUALS the
shipped default. A published map that may drift from the code is worse than no
map, because it invites reliance it cannot support.

Two findings sent to gate-house, in history/2026-08-29-layer-model-v06-review.md:
§6.4 obligation 1 (no side effect without a decision record) contradicts
obligation 3 and §9.3, with ops-warden's blessed fail-open stance as the
instance; and §6.4 mandates a stance-map inventory in §13 that §13 does not
implement — where ops-warden is currently the only PEP to have published one.

402 tests pass, ruff clean.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YWBMovyFoy9RRrfL7zKvPJ

Assistant: claude-code
Assistant-Model: opus
Assistant-Process: 4014535@bnt-lap001
Assistant-Session: d0036016-73e8-4da1-8e47-563e3ab39a3c
2026-08-29 10:20:49 +02:00

128 lines
5.5 KiB
YAML

# ops-warden — NetKingdom security layer declaration
#
# Framework: net-kingdom/canon/standards/security-layer-model_v0.4.md
# Assent: docs/adr/ADR-0010 (ops-warden's own voice, per §11 "who must declare")
# Validate: python3 scripts/check_layer_conformance.py
#
# §11 makes one check mechanical: "every direct Tooling client in a Staff
# repository maps to a declared §5.1, §5.2, or §5.3 entry". This file is that
# map. It is machine-readable because §5.3 requires it to be — ops-warden
# proposed that shape and is implementing it rather than declaring in prose.
#
# Conformance rule inherited from tenancy.yaml: accuracy, not altitude. A
# declared gap is TRACKED NON-CONFORMANCE (§11), never a claim of conformance.
schema_version: "0.1"
framework: netkingdom-security-layer-model
standard_version: "0.4"
repository: ops-warden
layer: staff
declared_by: docs/adr/ADR-0010
declared_at: "2026-08-29"
# §6.4 — ops-warden is PEP-shaped (it causes a protected side effect: issuing a
# certificate). Its unreachable-engine stance map is published separately, and
# asserted equal to shipped behaviour by tests/test_layer_conformance.py.
pep_stance: pep-stance.yaml
# Every direct contact with a Tooling-layer system (§4), one entry each.
tooling_contacts:
- id: ssh-ca-signing-write
shape: "5.3" # declared engine gap
module: src/warden/vault.py
symbol: VaultCA.sign
tooling: OpenBao
operation: "HTTP POST <addr>/v1/<mount>/sign/<role> with X-Vault-Token"
write: true
capability: "Sign a short-lived SSH certificate for an adm/agt/atm actor"
intended_owner: secrets-engine
blocked_on: >-
No engine exposes an SSH certificate signing surface. secrets-engine owns
credential abstraction, custody and lifecycle, which is the layer this
belongs in, but fronts no SSH-CA API today.
review: "2026-11-28"
note: >-
Production-verified and in daily use. This is the one lane ops-warden owns
permanently (§4). Signing continues while the gap is open: refusing would
remove production host access to close a documentation gap.
- id: desk-paste-once-provision
shape: "5.3"
module: src/warden/desk.py
symbol: _provision_to_openbao
tooling: OpenBao
operation: "bao kv put <path> <field>=- (value on stdin, never argv)"
write: true
capability: "Founder paste-once provisioning of a secret straight into OpenBao"
intended_owner: secrets-engine
blocked_on: >-
No engine exposes an attended provisioning surface for a value the founder
holds and no automated path can produce.
review: "2026-11-28"
note: >-
Attended and founder-operated (WP-0029). The value reaches OpenBao without
passing through a terminal, an argv, or the audit log.
- id: taint-metadata-read
shape: "5.1" # read-only diagnostic observation
module: src/warden/taint.py
symbol: fetch_taint_status
tooling: OpenBao
operation: "bao kv metadata get -format=json <path>"
write: false
capability: "Report EXPOSED taint (custom_metadata) without reading secret data"
intended_owner: secrets-engine
blocked_on: >-
No engine exposes a disclosure-taint query. Metadata-only by construction —
reading the data would be the 2026-07-16 vector this exists to avoid.
review: "2026-11-28"
- id: access-proxy-conduit
shape: "5.2" # conduit
module: src/warden/proxy.py
symbol: proxy_fetch, proxy_attended_login_exec
tooling: OpenBao, key-cape
operation: "Runs the catalog-declared owner fetch_command as a child process"
write: false
capability: "warden access --fetch/--exec/--out/--wrap for exec_capable lanes"
supplied_authority: none
evidence:
no_own_credential: src/warden/proxy.py::_caller_env
test: tests/test_proxy.py::test_conduit_supplies_no_authority_of_its_own
audit: "audit.jsonl records the caller, the lane, and the outcome; never a value"
note: >-
The §5.2 test is the supplied-authority property: ops-warden presents no
credential of its own, cannot widen what the caller could already do, and
the action reconstructs as the caller's. Governed by ADR-0002.
- id: caller-identity-token
shape: "5.2"
module: src/warden/caller_identity.py
symbol: resolve_caller_token
tooling: OpenBao
operation: "Runs the operator-configured caller_auth command, or reads token env"
write: false
capability: "Establish the caller's own identity for the pre-sign policy gate"
supplied_authority: none
detection: voluntary # runs an operator-configured command, so no fixed
# argv shape to scan for; declared rather than omitted
note: >-
Obtains the CALLER's credential by the operator's configured means; adds no
authority. Never mints, and never persists what it resolves.
# Contacts that are deliberately NOT Tooling contacts, recorded so the check is
# total rather than silently selective.
non_tooling_clients:
- module: src/warden/policy.py
target: access-engine (flex-auth)
rationale: "Engine API — §5 permits it; this is the shape §5 prescribes."
- module: src/warden/worker.py
target: state-hub, llm-connect
rationale: >-
Not catalogued in §4. The layer catalog scopes the security estate, and
neither the State Hub nor llm-connect appears in it, so no §5 shape applies
on the standard's own terms. Raised with gate-house 2026-08-29 as a scope
question rather than resolved unilaterally — see the assessment note.