docs(skills): record bearer staging rollout

This commit is contained in:
Jinwoo-H
2026-08-12 01:10:47 -04:00
parent 5a05992e26
commit 56e786fffc
2 changed files with 64 additions and 35 deletions
@@ -6,15 +6,16 @@ Last updated: 2026-08-12.
Implementation baselines captured by this checklist update:
- Orca implementation: `skills-share` (bearer-link changes follow baseline `6cff2aa507`; no PR).
- Orca Cloud: isolated `feat/skill-bundles` branch (bearer-link changes follow baseline `d7d1026`).
Production remains untouched.
- Orca implementation: `skills-share` at `5a05992e26`; no PR.
- Orca Cloud: PR `#320` merged to `main` as `0579cc1a71`; production remains untouched.
Validated so far:
- Final bearer-link changes pass 143 Orca Cloud API tests with one opt-in skip; 58 focused Orca
tests; four mixed-version wire tests; desktop/node/web typecheck; localization catalog,
extraction, and coverage; changed-code quality; max-lines; and diff checks.
- The final Cloud change passed PR run `31564069382` and merged-`main` run `31564235724`, including
the full monorepo build, lint, typecheck, tests, Terraform format, and Terraform validation.
- Local Node and web typechecks, changed-code quality gates, 94 skill-domain files with 770 tests
passed and 3 skipped, 134 Orca Cloud API tests with one opt-in integration skip, the full Cloud
monorepo test/typecheck/lint/build gates, and isolated Terraform formatting and validation.
@@ -95,15 +96,32 @@ Validated so far:
`31536141357`, merged as `4e91cf9`, and passed `main` run `31536329459`. The targeted Auth/API
plan then reported zero intended changes. Known SQL and artifact-bucket drift was excluded and
never applied.
- Cloud PR `#320` passed both required checks and merged as `0579cc1a71`. Targeted Terraform added
only `google_logging_project_exclusion.skill_share_bearer_request_urls` in staging: one addition,
zero changes, and zero deletions. This prevents per-link Cloud Run request URLs from entering
log storage while retaining route-template application telemetry.
- Guarded wake run `31564370807` prepared the reviewed staging topology. Deploy run `31564803943`
promoted `orca-cloud-api-staging-00051-yoq` at 100% traffic with immutable image digest
`sha256:511c0196511d2079bd9138092ad3cf065304b46a089b02d8df9318e0ae2e656a` and reported
`authenticatedSmoke: true` and `skillSmoke: true`.
- Candidate and canonical smoke produced 60 privacy-safe route-template events covering owner
upload/finalize/share creation/inventory/revocation and anonymous pinned/latest preview,
local/remote download grants, version selection, expiry, uniform unavailable responses,
package deletion, and object cleanup. Anonymous requests with no Authorization header and an
invalid header both returned the same `404 skill_share_not_found` response with `no-store`.
The serving revision had zero error logs and Cloud Logging retained zero per-link platform
request logs.
- Guarded sleep run `31546908228` stopped all three Relay cells after the physical staging work.
Independent GCP reads verified Cloud SQL policy `NEVER`, every MIG target at zero, and both
managed Cloud Run minimums at zero.
- Final guarded sleep run `31565009141` restored the same low-cost state after the bearer rollout;
independent reads found Cloud SQL `NEVER`/`STOPPED` and C1, C2, and C3 at target size zero.
Rollout gate: staging infrastructure and OIDC owner identity exchange passed. The earlier
authenticated skill lifecycle smoke predates bearer-link semantics and is not acceptance evidence
for anonymous recipient access. The shared staging data plane is back in its low-cost asleep state.
Production remains untouched; the anonymous bearer lifecycle, physical staging install journeys,
load, quarantine lifecycle, and soft-delete recovery gates remain.
Rollout gate: staging infrastructure, OIDC owner identity exchange, anonymous bearer lifecycle,
owner management, privacy-safe logging, and guarded rollback-capable deployment passed. The shared
staging data plane is back in its low-cost asleep state. Production remains untouched; the final
physical staging install journeys, load, quarantine lifecycle, and soft-delete recovery gates
remain.
Source plan: [Agent skill sharing and installation plan](./agent-skill-sharing-installation-plan.md).
@@ -1150,19 +1168,21 @@ still require the listed split metrics, latency panels, and alerts.
- [x] Apply reviewed Terraform and migrations to staging.
- [x] Deploy `orca-cloud-api` skill routes disabled by server-side flags.
- [x] Enable all four staging controls and pass the GitHub OIDC-authenticated artifact and skill
lifecycle smoke, including authorization, immutable versions, local/remote grants, rollback,
expiry, revocation, deletion, and object cleanup.
- [ ] Run upload/finalize/download tests for expiry, denial, oversize, corruption, deduplication,
revocation, deletion, and cleanup.
- [x] Enable all four staging controls and pass the GitHub OIDC owner plus anonymous bearer-link
lifecycle smoke, including owner management, immutable versions, local/remote grants,
rollback selection, expiry, revocation, deletion, and object cleanup.
- [x] Run staging upload/finalize/download smoke for anonymous denial, oversize, corruption,
expiry, revocation, deletion, and cleanup; retain tenant-deduplication coverage in the API
integration suite.
- [ ] Run desktop-to-local, paired-runtime, `windows 2`, WSL, and SSH journeys in staging.
The macOS desktop journey has live upload, finalize, share, and first-install evidence; its
exact-source update, rollback, revocation, removal, and deletion rerun remains.
- [ ] Give the live staging E2E an owner-private persisted test profile or a short-lived
non-interactive test credential so reruns do not open PKCE login tabs or copy a user's
primary Orca session.
- [x] Verify staging skill request logs contain only route, method, status, duration, and standard
process/request metadata, with zero credential or private-content matches.
- [x] Verify staging application logs contain only route templates, method, status, duration, and
standard process metadata, with zero credential or private-content matches. Exclude per-link
Cloud Run platform request logs through Terraform and verify zero are retained.
- [ ] Verify logs, metrics, traces, diagnostics, and support bundles contain no grants or private
package data.
- [ ] Load-test finalization and choose the fixed semaphore from observed memory, CPU, request
+29 -20
View File
@@ -53,16 +53,14 @@ that larger package-manager surface.
## Current execution status
The Orca implementation is on `skills-share`; bearer-link changes follow baseline `6cff2aa507` and
no Orca pull request exists. Cloud bundle ingestion and bearer-link work are isolated on
`feat/skill-bundles` after baseline `d7d1026`. The authenticated OIDC smoke landed in
`stablyai/orca-cloud#313`; the narrow Auth release-metadata convergence
follow-up landed in `#314`; the finalization and lifecycle fixes landed in `#317`. Staging
infrastructure exists, but its earlier authenticated recipient smoke predates the anonymous
bearer-link model and must be replaced before release. Production remains untouched. Local,
Windows, WSL, paired-runtime, and
Docker-backed SSH validation are substantially complete; the implementation checklist records the
exact evidence and remaining physical-host and failure-recovery gates.
The Orca implementation is on `skills-share` at `5a05992e26`; no Orca pull request exists. Cloud
bundle ingestion and bearer-link work merged through `stablyai/orca-cloud#320` as `0579cc1a71`.
The authenticated OIDC smoke landed in `#313`; the narrow Auth release-metadata convergence
follow-up landed in `#314`; the finalization and lifecycle fixes landed in `#317`. The final
anonymous bearer-link candidate and canonical smoke now pass in staging. Production remains
untouched. Local, Windows, WSL, paired-runtime, and Docker-backed SSH validation are substantially
complete; the implementation checklist records the exact evidence and remaining physical-host and
failure-recovery gates.
The local redesign now has an independently implemented Agent Plugins 1.0.0 skills-only bundle
manifest, deterministic one-or-many archive creation, bounded bundle extraction, additive bundle
@@ -75,12 +73,11 @@ bundle relay method with the same direct-download and staged-upload fallback; ol
rejected before transfer. The original single-skill path remains active for legacy shares while
bundle management migrates.
Cloud bundle validation is pushed on isolated branch `feat/skill-bundles` at `d7d1026`. It accepts
both envelopes during migration and stores the detailed bundle manifest with the immutable
version. Local bearer-link changes make active share IDs the recipient credential, keep owner
operations authenticated, retain legacy ACL rows only as migration storage, and prevent concrete
share URLs from entering request logs. It is not merged or deployed until the final gates and
anonymous staging smoke pass.
Cloud bundle ingestion accepts both envelopes during migration and stores the detailed bundle
manifest with the immutable version. Active share IDs are the recipient credential, owner
operations remain authenticated, and legacy ACL rows remain only as migration storage. A
Terraform-managed Cloud Logging exclusion prevents per-link Cloud Run request URLs from being
stored while privacy-safe route-template telemetry remains available.
The latest resilience pass adds bounded convergence after a lost final install response for both
paired runtimes and SSH. A staged retry creates a new upload instead of reusing a consumed upload.
@@ -111,8 +108,20 @@ promoted `orca-cloud-api-staging-00042-hef` with `authenticatedSmoke: true` and
The ten-minute, non-refreshable smoke principals exercised artifact CRUD and skill upload,
finalize, immutable versions, recipient/outsider authorization, local and remote grants, rollback,
expiry, revocation, deletion, and signed-object cleanup. Structured skill logs exposed only
privacy-safe request metadata and produced zero sensitive-value matches. Remaining staging gates
are end-user install journeys, load, quarantine lifecycle deletion, and soft-delete recovery.
privacy-safe request metadata and produced zero sensitive-value matches.
Cloud PR `#320` passed PR run `31564069382`, merged-`main` run `31564235724`, and a targeted
staging logging-exclusion plan of one addition, zero changes, and zero deletions. Guarded wake run
`31564370807` prepared staging. Deploy run `31564803943` promoted
`orca-cloud-api-staging-00051-yoq` at 100% traffic with immutable image digest
`sha256:511c0196511d2079bd9138092ad3cf065304b46a089b02d8df9318e0ae2e656a`. Candidate and canonical
smoke covered owner upload/finalize/share creation/inventory/revocation and anonymous preview,
local/remote download grants, version selection, expiry, uniform unavailable responses, package
deletion, and cleanup. Requests with no Authorization header and an invalid header returned the
same non-cached `404`; the serving revision produced zero errors, and Cloud Logging retained zero
per-link platform request logs. Guarded sleep run `31565009141` restored SQL to `NEVER`/`STOPPED`
and all Relay MIG targets to zero. Remaining staging gates are end-user install journeys, load,
quarantine lifecycle deletion, and soft-delete recovery.
## Research baseline
@@ -182,8 +191,8 @@ attribution. Revisit this decision if implementation work ever proposes copying
permissions.
- Requiring a provider-native plugin installer; unsupported agents still receive loose skills.
- End-to-end encryption from Orca Cloud operators in the initial trust model. Private means
access-controlled to authorized Orca users and organizations. Application-level package
encryption can be added later without changing package identity.
unlisted and accessible through an active bearer link, with authenticated owner management.
Application-level package encryption can be added later without changing package identity.
## Product experience