feat(api): extend mutation replay through gateway interceptors

[mrunalp]

🏗️ build-from-issue-agent

Summary

Extend durable request admission and replay to the remaining 24 ordinary user-callable unary mutations. Typed receipts preserve operation outcomes without storing credential-bearing request or response snapshots, and intercepted retries remain subject to current validation.

Related Issue

Part of #3051, phase 3b. Stacked on #3321 (feat/3051-mutation-replay/mrunalp). This PR does not close the issue: credential capabilities, exec/streaming semantics, curated SDK request-ID helpers, and the final mutation audit remain phase 4.

Changes

• Add optional UUID request IDs to sandbox lifecycle/attachment, service, provider/refresh, profile, and policy/config mutations; regenerate Go bindings and migrate Rust request literals.
• Capture identities from actual handler reads/writes. Sandbox replay loads the live original UUID; provider/profile/service projections require recorded versions; refresh projections require the original grant epoch; policy/config and delete receipts preserve historical outcomes.
• Bound completed receipts to 64 KiB and leave uncertain admissions permanently unresolved. Protect potentially sensitive payload fingerprints with domain-separated HMAC derived from existing JWT or primary TLS private material. Missing or rotated keys fail closed, and replicas must share key material.
• Carry original client payloads only in private in-memory middleware extensions, freeze request IDs, reauthorize original/effective scopes, rerun interceptor modification/validation, and suppress duplicate post-commit observation. Observation remains best-effort, not a durable outbox.
• Add 14 server regression tests, extend SQLite/PostgreSQL restart/concurrency coverage to keyed adapters, and add a live sandbox lifecycle/replacement regression.
• Update architecture/gateway.md and docs/reference/api-errors.mdx with the supported RPCs, replay projections, key-material requirements, and remaining exclusions.

Deviations from Plan

No scope expansion. The reviewed phase-3b design uses method-specific receipts and existing private gateway material rather than generic response persistence or new deployment configuration.

Testing

• mise run pre-commit passes.
• Targeted replay suite: 25 passed; opt-in PostgreSQL scenario passed separately.
• TypeScript SDK CI: 108 tests passed, including codegen/type checks/build.
• Documentation validation: 0 errors, 3 existing warnings.
• Full mise run ci: 1,546 server unit tests passed (9 existing ignored); repository tests and Go SDK CI passed; Python 247 passed, TypeScript 108 passed.
• Docker conformance smoke and both Rust workspace lifecycle tests passed.
• Python sandbox/workspace API E2E: 11 passed, 159 intentionally deselected, including the new replay regression.

Local Rust verification disables an unusable sccache wrapper via a task-shell override. Docker supervisor staging uses its supported OPENSHELL_AUDITABLE=1 build mode, which removes the wrapper after nested mise setup. No checks are skipped and repository tool configuration is unchanged. Disposable E2E gateways, sandboxes, networks, and PostgreSQL test resources were removed.

Checklist

• Follows Conventional Commits.
• Commit is signed off (DCO).
• Independent implementation review found no correctness or security blockers.
• Reviewed related skills with the maintenance map; no skill inventory, workflow, or instruction changes required.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[mrunalp]

🏗️ build-from-issue-agent

E2E Test Attestation

Local Docker-backed E2E passed against the contents committed as 80415d1604988cf4fcfcfef8f13660fbe57232be.

Commands:

OPENSHELL_AUDITABLE=1 OPENSHELL_E2E_DOCKER_TEST=workspace_lifecycle mise run --shell 'env RUSTC_WRAPPER= bash -c -o errexit -o pipefail' e2e:docker
OPENSHELL_AUDITABLE=1 E2E_PARALLEL=0 PYTEST_ADDOPTS="-k 'test_sandbox_api or test_workspace_api' -v" mise run --shell 'env RUSTC_WRAPPER= bash -c -o errexit -o pipefail' e2e:python

Both commands created isolated standalone gateways with the Docker compute driver, current gateway/CLI binaries, and a freshly built static supervisor. The auditable supervisor build mode and task shell avoid a local sccache execution error; no repository checks were disabled. The initial Docker attempt stopped during supervisor staging before tests started. Both final commands passed, and their gateways, sandboxes, and temporary networks were removed.

Results

• Standalone CLI conformance smoke: passed (status, create, exec, deletion).
• Rust: 2 passed; 0 failed; 0 ignored.
• Python: 11 passed, 159 deselected in 23.46s. The deselected tests were outside the focused sandbox/workspace API modules; no selected test was skipped.

Tests Executed

• e2e/rust/tests/workspace_lifecycle.rs::workspace_terminating_rejects_creates — passed.
• e2e/rust/tests/workspace_lifecycle.rs::workspace_full_crud_lifecycle — passed.
• e2e/python/test_sandbox_api.py::test_mutation_replay_preserves_sandbox_lifecycle_and_replacement — passed.
• e2e/python/test_sandbox_api.py::test_sandbox_api_crud_and_exec — passed.
• e2e/python/test_sandbox_api.py::test_sandbox_interactive_exec_honors_tty — passed.
• e2e/python/test_sandbox_api.py::test_list_scoped_and_for_all_workspaces — passed.
• e2e/python/test_sandbox_api.py::test_sandbox_labels_and_selectors — passed.
• e2e/python/test_workspace_api.py::test_workspace_crud — passed.
• e2e/python/test_workspace_api.py::test_workspace_create_with_labels — passed.
• e2e/python/test_workspace_api.py::test_workspace_list_includes_created — passed.
• e2e/python/test_workspace_api.py::test_workspace_delete_nonexistent_raises_not_found — passed.
• e2e/python/test_workspace_api.py::test_workspace_get_nonexistent_raises_not_found — passed.
• e2e/python/test_workspace_api.py::test_workspace_request_id_replays_without_deleting_replacement — passed.

The new replay regression verifies create/stop/start/config/delete results and server replay metadata across actual sandbox reconciliation, then recreates the same name and proves delete replay preserves the replacement.

[github-actions]

🌿 Preview your docs: https://nvidia-preview-pr-3323.docs.buildwithfern.com/openshell

[johntmyers]

gator-agent

PR Review Status

This accepted issue-backed phase is project-valid, and the full initial code review found no blocking defects. The user-visible replay contract is documented; required current-head testing must now be dispatched before pipeline monitoring begins.

Blocking findings:

• No blocking findings remain

Carried findings:

• None

Non-blocking suggestions:

• None

Gator metadata

• Validation: Implements accepted issue #3051 phase 3b on its reviewed stacked base #3321.
• Docs: Replay behavior and API errors are updated in architecture/gateway.md and docs/reference/api-errors.mdx; navigation changes are not needed for the existing reference page.
• Checks: Current-head Branch Checks and Helm Lint have not run yet; DCO, vouch, Trivy, docs preview, and available synthetic gates are green.
• E2E: test:e2e is required for sandbox lifecycle, gateway/interceptor, policy/config, and provider-flow changes and is being dispatched.
• Head SHA: 80415d1604988cf4fcfcfef8f13660fbe57232be
• Base SHA: 5dc3c72940b13ca0da55a0b62ec792d71e50cf17
• Merge base SHA: 5dc3c72940b13ca0da55a0b62ec792d71e50cf17
• Patch ID: 132f907ff26a660e8d851f936f50f356109b090a
• Gator payload: 9
• Review mode: initial
• Previous reviewed SHA: none
• Review budget exhausted: no
• Maintainer decision required: no
• Next state: gator:in-review

[github-actions]

Label test:e2e applied, but pull-request/3323 does not exist yet. A maintainer needs to comment /ok to test 80415d1604988cf4fcfcfef8f13660fbe57232be to mirror this PR. Once the mirror exists, re-apply the label or re-run Branch E2E Checks from the Actions tab.

[johntmyers]

/ok to test 80415d1

[johntmyers]

gator-agent

Monitoring Complete

Monitoring is complete because this PR has merged.

Final status: The PR reached gator:merge-ready with maintainer approval and required checks complete.

I removed the active gator:* label because there is nothing left for gator to monitor on this PR.

Gator metadata

• Head SHA: 80415d1604988cf4fcfcfef8f13660fbe57232be
• Gator payload: 9