Record published consumer build and refine private input inverse and PDP work
All checks were successful
CI Smoke / host-smoke (push) Successful in 0s
CI Smoke / container-smoke (push) Successful in 2s

Assistant: codex
Assistant-Model: gpt-6.1-sol
Assistant-Session: 01a1468b-63fa-7710-85dc-8f2e16b3ebfa
This commit is contained in:
tegwick 2026-10-18 00:13:34 +02:00
parent 201e2fa0f9
commit b3f31d9f26

View file

@ -1788,3 +1788,106 @@ owner boundary resolved; reference-only source does not prove it. Native go,
deployment and independent browser/group/revocation observations remain existing
T06/T16 and owner gates. No credentials, image deployment or new kernel step
is authorized by this task.
## T40 — Publish the independently reviewed full consumer build path
```task
id: NK-WP-0048-T40
status: done
depends_on: [NK-WP-0048-T38]
priority: high
state_hub_task_id: "7cc93671-fe94-5108-a486-1b0029e69487"
```
Completed 2026-10-18, release source/tooling outcome extracted from T39.
Apps RAPPS-WP-0016-T06 owns the pinned official-source/frontend/bindata/static
binary/OCI recipe. The original profile produced identical complete binaries and
local OCI archives. Independent review rehashed those artifacts, re-extracted all
2,234 embedded files and found a host process argv visibility defect through
/proc. The corrected recipe adds a private PID namespace; a synthetic unrelated
process canary is visible with r1 and denied with exact r2. Existing full builds
are historical r1-profile evidence, not an executed full r2 build claim.
Canonical Apps source c29408ae contains the six reviewed build paths plus its
work-record refinement; the normal owner projection receipt is in Custodian
apps-release-source-integration-r1. Exact candidate patch sha256:b7ad8f8fd908532800f4e7eb0f7ad85d77e21a0b729cb5115ade3d5fe3e4acd8
and review sha256:6f016d53a7ec9ece2f9da9686e76127c52d17066f2b3d371628832ac9537ac5c
are retained with original changes-required review and harness refinements.
No image/default values, registry publication, formal owner assent or native
acceptance changed. T39 retains integrated successor package/image provenance,
architecture/owner dispositions and bounded deployment/inverse readiness.
## T41 — Integrate the real private-input CLI and chart revision
```task
id: NK-WP-0048-T41
status: progress
depends_on: [NK-WP-0048-T38]
priority: high
state_hub_task_id: "6edf049f-2c52-5040-8d38-0bb1bc78fa31"
```
Owner: Apps RAPPS-WP-0016-T07. Current public-source implementation adds actual
Forgejo add/update --secret-file and adapts the exact chart helper plus fixed
CLIENT_SECRET read-only subPath reference. Next: independently review the exact
candidate, including real CLI/SQLite update preservation and refusals, descriptor
file replacement/permissions and actual Helm command/mount rendering. Preserve
legacy and three-group semantics and the current custody route/key. Custody,
mount/reader/version admission and rotation/restart remain runtime obligations;
no agent or public receipt receives the value. Re-freeze the accepted software/
chart/type/package only after review; T06 native proof remains separate.
Evidence under Custodian forgejo-private-input-r1; do not substitute a path flag
without suppressing the chart's legacy secret argument.
## T42 — Integrate the actual durable application-source transaction and inverse
```task
id: NK-WP-0048-T42
status: progress
priority: high
state_hub_task_id: "3d1b9128-bade-5bd9-8660-22a14eea2d37"
```
Owner: Apps RAPPS-WP-0016-T08; retained umbrella T34. Actual Forgejo model-layer
work compares complete raw persisted state privately and records intent/outcome
in the same DB transaction. SQLite write transactions and PostgreSQL table locks
must exclude competing DB writes; unsupported databases refuse. Tests include
real SQLite competing connections and process crash/restart. Fixed SQL-error,
slow-query and trace diagnostics must not leak private journal arguments.
Next: independently review exact primitive/schema/logging/recovery code, then
bind it through authenticated current operation/inverse sessions and admitted
evidence/release serialization. The DB primitive itself grants no authority and
is not a complete operation backend. Its private journal introduces credential
copies in the owner's DB: track reader custody, capacity, retention and recovery;
never erase uncertainty or downgrade while a linked source prevents safe removal.
All application/Helm writers and current provider state remain T34 obligations.
Evidence under Custodian apps-operation-backend-r1; no private production DB read.
## T43 — Bind the consumer and protected operations to the actual single PDP
```task
id: NK-WP-0048-T43
status: progress
priority: high
state_hub_task_id: "ac77b046-1226-5e5a-a212-512cb4fdc484"
```
Owners: Access Engine for producer adapter/policy; Apps RAPPS-WP-0016-T09 for
consumer enforcement; retained umbrella T37. Next: implement and independently
review the producer-owned standalone Go adapter using exact pinned API/signing
source, current authenticated request/policy/registry/caller bindings and bounded
nonce/lifetime checks. The actual provider must validate issuer JWT signatures,
algorithm/audience/type/lifetime and retain a trusted identity snapshot separate
from mutable UserInfo groups/claims before calling the PDP. Do not treat decoded
Goth RawData as verified identity or let UserInfo overwrite issuer/tenant/assurance.
Any producer numeric-decoding fix needs actual producer/consumer parity tests.
Keep the accepted three-group any-of set as PDP policy input; do not invent an
issuer verdict or allow a local compatibility gate to replace the PDP. A callback
check proves login only: actual HTTP/API/Git/SSH resource/action and session/
revocation coverage must match the final application profile. No source adapter,
synthetic key, stale cached decision or Boolean callback constitutes authority.
Evidence under Custodian apps-access-engine-integration-r1; live source/key/caller/
policy admission and full profile acceptance remain open under T06/T16/T37.