feat(agents): file new Solo chats in explicit personal workspaces - #2857

Summary

• Add solo_workspace_list and optional workspace on new private Solo creation, for locally hosted agent runs only. Require the host-bound owner explicitly in members and an exact existing personal workspace ID.
• Reuse the existing encrypted Solo-preference/channel-kit writer. Preserve agent signing/creator, membership and existing no-workspace behavior; do not copy workspace defaults or Canvas.
• Bind placement to each exact future chat ID with one-use creation provenance, including overlapping creates. Preserve the created ID and distinguish known pre-assignment refusal from uncertain completion after a write starts; never retry creation as recovery.
• Do not add an arbitrary move operation, external/standalone capability, Nexus-worker support, or plugin-handle authority.

Validation — full suite is NOT green

Original reviewed head: bb37256a8dcf8a6b8997a7788d9eb6008f029f06 (independent workspace PR; picker ordering is separate).

• Exact five complete affected test files: 734 passed, 1 existing skip; project type check and mandatory commit hooks passed.
• Last exact full Vitest run: 996 files / 18,858 tests passed; 1 file / 1 test failed; 8 files / 22 tests skipped (1127.61s). Sole failure: browser-host/vite-config.test.mjs — config-loader Node subprocess hit its existing 30-second ETIMEDOUT limit. Root cause is not established. No timeout/assertion weakening and no further blind full-suite rerun.
• Earlier full-run App.profile and WorkflowLanding timeouts passed as complete files in isolation and did not recur in the final run. Those isolated passes do not erase the final non-green result.
• Deterministic real-runner overlap regression fails when shared provenance-map clearing is restored. Gated post-durable-write regression fails when uncertainty is incorrectly reported as definitive not-filed. Coverage includes actual preference-writer cancellation after fixture save and app-client retirement after completed placement.
• Independent re-review: PASS, quality 9/10, after one completed fix/re-review cycle addressing overlap provenance and truthful uncertain outcomes.

Native and human exercise

Combined acceptance app at 1ef7c0a91361310a47a4c5fae4a173ddb12b1a8c contains this reviewed workspace implementation unchanged. It ran separately from installed Buzz with the existing owner identity and feedback community.

• Owner saved a separate Chat Fixes Local Acceptance Goose agent. Matching native snapshot shows it running locally (place: here); its session 20261106_83 used an MCP listener owned by the acceptance app, not installed Buzz.
• Recorded actual calls: solo_workspace_list resolved Test workspace, then channel_create supplied that exact workspace ID with solo: true, private visibility and explicit human + agent members. Result reported the created ID and filed in workspace.
• Matching native sidebar/preferences show chat 678377f5-f225-41f5-916c-42de83b50f64 directly in workspace 819d0711-bbfb-48c2-847e-67506e19621f; metadata identifies the local agent as creator and the human as member. The human reported “worked!!” and requested PR publication.
• The human-selected chat name was New chat, not the proposed distinctive test name. No duplicate was created to repair naming.
• Prior Bedford attempt ran through the older installed host. Its separate unassigned chat 45df41eb-b9c1-4fbd-8a85-a6e83e6041fe is preserved and is not this successful acceptance result.

Remaining native evidence: test-instance restart persistence was not exercised or explicitly reported; no independent encrypted-delivery receipt, raw signature audit, or no-message/Canvas/default-copy inspection is claimed. The native successful workflow is evidence of actual tool wiring and placement, not proof of all those boundaries. Unit coverage establishes the modeled safety contracts.

November 11 integration and review readiness

Current head: c72d3d1479038e89d10a0eaaa67158445dbc2f45, merging original accepted bb37256 with main 128cd4c2f38801d5f773d5d5ec95a1ea5594b374. Original commits and attribution are preserved; no history rewrite.

• Four conflict files reconciled. Upstream full-agent external access remains intact; remove only soloWorkspaces from the external run handle so workspace discovery/filing remains local-turn-only as agreed. Upstream external implementation/tests are unchanged. No UI redesign or unrelated CI repair.
• Complete five affected test files: 752 passed, 1 existing skip (35.37s), plus project typecheck. Tested tree 60dd337ddabde8c5ed17240f2dad09b79e98ea0a equals the committed tree. New external-workspace refusal regression fails before the exclusion and passes in the complete service file.
• Mandatory pre-push types/related tests passed (206.49s), design passed (7.31s); Clippy lane correctly skipped Rust compilation for this TS-only feature. Machine policy hooks passed. Exact local, remote and PR heads verified after push.
• Independent integration-delta review: PASS, 9/10, no blockers at c72d3d1. Existing local/native and human workflow evidence above is retained; local placement behavior is unchanged. No new native or browser campaign was run.
• Old CI at bb37256: JavaScript 3/6 timed out in unchanged App.profile.test.tsx; Vitest flake 1/2 first iteration timed out in an existing Attention polling case and had a follow-on fake-timer error, then passed iterations 2–5. No root cause claimed or unrelated repair added. Fresh hosted CI is pending for c72d3d1; old failures are not described as green.

Human authorized resumption, conflict reconciliation, publication and readiness on November 11. Required hosted checks and reviewer/code-owner approval still gate merge. No merge, release, installation or artifact cleanup is included. Historical full-suite and native evidence limitations above remain disclosed. No browser journeys added or removed.

November 11 — Kalvin P3 coverage follow-up

Current published head: 2c3af95; prior reviewed/accepted integration head c72d3d1479038e89d10a0eaaa67158445dbc2f45.

• Test-only follow-up adds cached: true and writable: false independently to the existing unconfirmed-authority table. Both require list/prepare refusal, with no relay-evidence read or assignment. Existing valid authority/workspace setup is retained.
• Guard-removal mutation (only state.cached || !preferences.writable removed): both new cases fail, resolving Kalvin's reported coverage gap. Restored production, complete service test file: 538 passed / 1 existing skip, 35.50s.
• Production files are unchanged by this follow-up. Existing agent/human feature acceptance remains applicable; no new behavior or app-test campaign is claimed or required.
• Focused independent review: PASS, 9/10, no blockers; normal pre-commit hooks passed. Pre-push types/related tests 194.08s and design 6.97s passed; Rust checks not applicable to this test-only delta. Exact remote head verified, clean worktree.
• Fresh hosted checks for this successor head are pending; earlier hosted success applies only to c72d3d1. The PR remains ready for review; required human/code-owner approval is still pending. No merge authorized.