[teamleaderleo]

A session whose agent dies mid-turn can continue on the same conversation, using the existing bounded retry and Stop rules. Before restarting, acpmux recreates a recorded linked worktree from its local branch, origin branch, or last commit. If recovery fails, cwd stays unchanged and the prompt receives the existing retryable session.folder_missing refusal.

Leo's exact words: "ACPMUX daemons, peer routing and messaging across machines."

The checkout uses the existing configuration-isolated path, with hooks and repository filters disabled. Remote-origin sessions do not recreate worktrees. Usage-limit and sign-in failures are never automatically continued; stderr is drained before classifying a process exit.

This is only the dead-agent/worktree item. Incoming prompts retain the existing transport-continuation policy; this PR adds no idle-status or parent-folder-routing change. No UI surface changed, so there are no gallery variants or screenshot URLs.

Validation

• Final focused hosted verification run 40043044644 on 86055a738622d20a1e5d59b32b9544f00e1b9c6b: Linux ACPMUX tests, MSRV, artifact build, and Go checks passed; Linux clippy/panic-ratchet failures are unchanged base findings, and the macOS job failed while acquiring its fleet runner.
• Rust god-file check passes against integrated base ec475c71f177e4bb56e4163a25b3c7a8640c7e56.
• verify-local.py: 10 checks pass; localization parity debt also exists in the unchanged base inputs.
• pr-overlap 23120 inspected overlaps acpmux: forward a Claude Code chat's plan usage limits to clients #24293, cmux NEXT Inspector Connectors: a failed MCP server has Reconnect #24272, acpmux: Claude Code's /context breakdown and MCP reconnect #24223, cmux-next trust: Not now can leave a restricted Claude Code chat instead of a dead end (setting, off by default) #23200 and cmux NEXT sidebar: rows keep their places in the session tree, with a stable/attention toggle #21944: shared agent/session plumbing; this is the existing owner PR for dead-agent/worktree recovery.

Changelog

Fixed: Agent chats resume after an agent process exits mid-turn and can recreate their deleted linked worktree before restarting.

Bead: cx-fbt7f. Owner stop condition: merged.

Agent session: huge-blue

[teamleaderleo]

Owned by projects-a4 on cosmic-gray · session

[teamleaderleo]

Owned by cx4-au23120 on huge-blue

[teamleaderleo]

Bead cx-fbt7f created for this PR by the cmux coordinator.

[blacksmith-sh]

Found 14 test failures on Blacksmith runners:

Failures

Test	View Logs
/tmp/tmp20wcsegk/summary-fail.json (live):/summary-fail.json (live):	View Logs
/tmp/tmpp_iukoua/summary-missing.json (live):/summary-missing.json (live):	View Logs
[tabs] debug.tab_switch timed 0 of 40 switches/ [tabs] debug.tab_switch timed 0 of 40 switches	View Logs
[tabs] debug.tab_switch timed 0 of 40 switches/ [tabs] debug.tab_switch timed 0 of 40 switches	View Logs
[tabs] the window draws no frames/[tabs] the window draws no frames	View Logs
[tabs] the window draws no frames/[tabs] the window draws no frames	View Logs
convention violations found (ERROR lines above)./ convention violations found (ERROR lines above).	View Logs
test_a_pull_request_that_changes_the_shortcut_inputs_still_rewrites_them (main.Sett ingsOnlyArtifactLeavesTheActionFollowers.test_a_pull_request_that_changes_the_shortcut_ inputs_still_rewrites_them) (job='repair')/ test_a_pull_request_that_changes_the_shortcut_inputs_still_rewrites_them (main.Sett ingsOnlyArtifactLeavesTheActionFollowers.test_a_pull_request_that_changes_the_shortcut_ inputs_still_rewrites_them) (job='repair')	View Logs
test_base_passes (main.NoticesPairTests.test_base_passes)/ test_base_passes (main.NoticesPairTests.test_base_passes)	View Logs
test_every_routed_input_triggers_the_workflow (main.PullRequestTiers.test_every_rou ted_input_triggers_the_workflow) (path='.git/config')	View Logs

...and 4 more test failures. View all on Blacksmith: run 40043018826, run 40043018876

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[github-actions]

cmux status _{(one comment, edited in place; each line is updated by the job that owns it)}

Review 86055a7: HQ feed card · gallery before/after runs · checks

App of this head (app.zip, kept 1 day)

gh workflow run reload-build.yml -R manaflow-ai/cmux --ref feat-cmux-next -f tag=pr23120 -f ref=acpmux-revive-dead-agent -f platform=macos -f backend_mode=local

red: cmux-next checks (god files, concurrency, crash safety, l10n); no finished base run on 80beafb to compare; do not merge until green; see lane-rules

[teamleaderleo]

From the persistent-machines lane: I had this same item (turns surviving a harness restart) on my plan and dropped my version, so this PR is the one. Three cases my brief called for that I couldn't confirm from the description. They may already be covered:

1. Fallback profile. When turns.rs (~398) re-sends to a fallback profile after "agent process closed", does auto-resume stay off? A test that exactly one of the two fires would pin this down.
2. Planned stops. A planned restart or update (chat Idle or Closed) should never count as an agent death and trigger a continuation.
3. Retry cap. A test where the harness dies on every respawn and the retries stop after the cap, leaving the chat disconnected with the queue held.

No reply needed.

[teamleaderleo]

Fleet test results at head 146d2b628fe5e5ab3621983c4f92877e81ddcffd, on cmux16 (Darwin arm64):

• cargo fmt -p acpmux --check: clean
• cargo clippy -p acpmux --all-targets: no new warnings (the 4 it prints are in files this PR doesn't touch: cmux-worktree-care/store.rs, steer_now_tests.rs, self_command.rs, tests/prompt_images.rs)
• cargo test -p acpmux --lib: 370 passed, 2 failed. Both fail the same way on base 6b19b5d9180: protocol::tests::protocol_registry_names_every_served_method and worktree_sessions::care_tests::care_a_stopped_daemon_s_sessions_are_use_not_open.
• cargo test -p acpmux --test hub_integration (all of it): 319 passed, 2 failed, 1 ignored. Both fail the same way on base 6b19b5d9180: session_offers::an_offered_chat_is_a_chip_only_the_person_starts_and_it_reports_its_parent and transfer_guard::a_chat_with_no_asking_mode_moves_under_peer_control_but_not_under_web.
  • All 7 agent_revive::* tests pass, as do all 12 auto_resume::*, all 8 status_indicator::* and all of fork_through::*.
• scripts/cmux-next/check-no-godfiles.sh --only rust --base origin/feat-cmux-next: exit 0
• scripts/cmux-next/crash_ratchet.py: acpmux expect count back at 10, the same as base

CI reds at this head, none in this diff (acpmux Rust only):

• cmux-next checks: check-scrollbars (Swift/CSS), FIRST-TURN-COST (Chief), conventions ERROR in HomeThoughtHeadline.swift, and the package graph CmuxNextHomeEngineTests. All are base reds.
• Earlier heads also showed publish the merge tree, merge freeze and high-signal review link failing with a GitHub API 403 (rate limit), not a test.

[teamleaderleo]

codex-review: SHIP at 146d2b6

• Reviewed the current diff, its focused tests, and the affected call paths; no correctness bug found.

[cursor]

Bugbot needs on-demand usage enabled

Bugbot uses usage-based billing for this team and requires on-demand usage to be enabled.

A team admin can enable on-demand usage in the Cursor dashboard.

[cursor]

Bugbot needs on-demand usage enabled

Bugbot uses usage-based billing for this team and requires on-demand usage to be enabled.

A team admin can enable on-demand usage in the Cursor dashboard.

[teamleaderleo]

Review (Codex gpt-6.1-sol xhigh): SHIP at 86055a7

Independent final-head review covered dead-agent continuation, linked-worktree recreation and identity checks, archive detached restoration, branch-tip races before exposure, and detached replacement safety. The final focused fleet run passed Linux ACPMUX tests, Rust MSRV, artifact build, and Go checks; remaining Linux lint reds are unchanged base findings, and the macOS job failed during fleet runner acquisition.

[github-actions]

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4005909948001 (66747a14cb)
• cmux NEXT nightly 1.0.0-nightly.4005999011801 (09af538c30)