Restore native local child owners in resumed cloud parent drivers

[warp-agent-staging]

Description

Restore native/Oz local child owners inside a resumed cloud parent’s existing agent-driver process. Strict paginated discovery completes before parent execution; each eligible child gets a hidden owner and its own wake listener without fetching its transcript eagerly.

A wake or pane open coalesces transcript hydration into the existing conversation ID. Normal inbox delivery stays gated until installation succeeds; setup or hydration failure marks that child Error. Desktop and Observer discovery stays passive for cloud-owned parents.

Review order: the approved contract in specs/QUALITY-2186/TECH.md, shared discovery, headless owner bootstrap, then the streamer/controller hydration gate and passive seeding. Ownership uses the existing PaneGroup rather than a new workspace. Existing parent handoff/credential safeguards apply; no additional physical-VM teardown barrier is introduced.

Companion server: https://github.com/warpdotdev/warp-server/pull/20288. Deploy it before enabling the default-off RestoreLocalChildOwners flag in any environment; an older server’s valid-but-wrong location cannot be detected by this client.

Linked Issue

QUALITY-2186

• The linked issue is labeled ready-to-spec or ready-to-implement. This is a Linear issue; the GitHub issue label does not apply.
• Where appropriate, screenshots or a short video of the implementation are included below. The requester explicitly declined UI verification.

Testing

Passed before final formatting:

• cargo nextest run -p warp --lib -E 'test(restored_local_child_)': 9 cases covering strict traversal/decoding, classification, passive Observer seed/cursor, owner promotion, hydration coalescing, and transcript Error handling.
• 148 adjacent streamer/pane/card/icon tests. After merging current master, all 24 selected restoration, terminal-driver, and prewarmed-driver tests passed. The mock-only run used env -u WARP_WORKLOAD_TOKEN WARP_ISOLATION_PLATFORM=docker to avoid unavailable Namespace token issuance in this sandbox.
• cargo check -p warp and cargo clippy -p warp --all-targets --tests -- -D warnings.
• Server handler/integration suites, Go lint/vet, and full Agents/Factory web suites passed on the companion branch.

Outstanding approved-spec validation: no live VM A→B handoff or actual wake→replay→child message-turn run was exercised. Component tests do not prove a driver-level pagination barrier before initial or --skip-initial-turn execution, successful post-hydration replay/turn, or every failure-stage variant. These drafts remain blocked pending that proof; this is not a fully validated handoff.

• I have manually tested my changes locally with ./script/run. Not run; UI verification was declined. The headless cross-VM/message path still needs integration proof.

Agent Mode

• Warp Agent Mode - This PR was created via Warp's AI Agent Mode

Plans:

• Implement approved QUALITY-2186 local-child restoration

[warp-agent-staging]

This PR was generated with Warp.

Comment @warp-staging-factory on this PR to send it follow-up work.

  

[warp-agent-staging]

Overview

The paired drafts restore native local child owners and gate inbox delivery on transcript hydration. Required headless integration proof is still missing.

Concerns

• Important — required integration proof is incomplete. specs/QUALITY-2186/TECH.md requires VM A→B restoration, a real wake→replay→child turn, and pagination/skip-initial-turn barriers; the passing component tests do not exercise these paths. Which disposable paired client/server environment should run that proof before these drafts are treated as ready?

Verdict

Checks: build pass, component tests pass, CI red (draft gate intentionally skips build/test jobs), visual proof n/a (declined).

Found: 0 critical, 1 important, 0 suggestions, 0 nits. Integration environment/proof decision required; no approval or denial submitted.

Responding as eng-factory-slim: Open session · View in factory