fix-tui-supervisor-coexistence — Workgraph live mirror

Metadata

Status	done
Assigned	`agent-1800`
Agent identity	`3184716484e6f0ea08bb13539daf07686ee79d440505f1fdf2de0357707034c3`
Model	`codex:gpt-5.5`
Created	2026-05-02T23:54:00.607718294+00:00
Started	2026-05-03T01:04:58.925894755+00:00
Completed	2026-05-03T01:22:36.419747852+00:00
Tags	`fix,nex,chat,tui,supervisor,bug`, `eval-scheduled`
Eval score	0.74
└ blocking impact	0.70
└ completeness	0.70
└ constraint fidelity	0.55
└ coordination overhead	0.75
└ correctness	0.75
└ downstream usability	0.70
└ efficiency	0.80
└ intent fidelity	0.69
└ style adherence	0.80

Description

Apply the chosen approach (single-owner sentinel OR cooperative-handoff observer-mode) from research-tui-supervisor-coexistence. Goal: when the user is connected via TUI, the supervisor stops trying to respawn its own nex handler.

Implement directly — do not decompose further.

File scope (limit to these files)

src/tui/viz_viewer/state.rs (around 13696-13753, takeover dance) — if the chosen approach writes a sentinel/flag here
src/commands/service/coordinator_agent.rs (pre-flight check at the top of the supervisor loop) — to consult the sentinel/flag
src/session_lock.rs (only if the protocol extends here)
tests/

DO NOT touch: event.rs (no input changes), nex.rs handler logic, agency code.

Validation

Failing test written first that exercises the coexistence boundary
Implementation makes the test pass
cargo build + cargo test pass with no regressions
Live verification: open TUI, switch to chat tab; assert supervisor stops respawning while TUI is connected; close TUI; assert supervisor resumes

## Description
Apply the chosen approach (single-owner sentinel OR cooperative-handoff observer-mode) from research-tui-supervisor-coexistence. Goal: when the user is connected via TUI, the supervisor stops trying to respawn its own nex handler.

Implement directly — do not decompose further.

## File scope (limit to these files)
- src/tui/viz_viewer/state.rs (around 13696-13753, takeover dance) — if the chosen approach writes a sentinel/flag here
- src/commands/service/coordinator_agent.rs (pre-flight check at the top of the supervisor loop) — to consult the sentinel/flag
- src/session_lock.rs (only if the protocol extends here)
- tests/

DO NOT touch: event.rs (no input changes), nex.rs handler logic, agency code.

## Validation
- [ ] Failing test written first that exercises the coexistence boundary
- [ ] Implementation makes the test pass
- [ ] cargo build + cargo test pass with no regressions
- [ ] Live verification: open TUI, switch to chat tab; assert supervisor stops respawning while TUI is connected; close TUI; assert supervisor resumes

Depends on

Required by

Log

2026-05-02T23:54:00.590704619+00:00 Task paused
2026-05-03T00:51:49.751467679+00:00 Task published
2026-05-03T01:03:11.094783548+00:00 Spawned by coordinator --executor codex --model gpt-5.5
2026-05-03T01:04:06.047943836+00:00 Evaluator assignment detected before actor completion: task is still in-progress with no commits/artifacts; wg evaluate run refused because status is InProgress.
2026-05-03T01:04:22.796026298+00:00 Task marked as incomplete (attempt #1 (2 remaining)): Evaluator role was assigned to the implementation task before any actor output existed. No implementation commits/artifacts are present; wg evaluate run refused because the task is still InProgress. Needs redispatch to a programmer actor, then evaluation after completion.
2026-05-03T01:04:58.925899153+00:00 Spawned by coordinator --executor codex --model gpt-5.5
2026-05-03T01:05:10.686459771+00:00 Evaluator resumed; checking task status and artifacts before grading.
2026-05-03T01:05:29.225511596+00:00 Proceeding with implementation inline: sentinel protocol in session_lock.rs, supervisor defer check, TUI writes/clears sentinel, focused tests.
2026-05-03T01:14:16.191962567+00:00 Focused regression validation passed: sentinel round-trip/stale tests and supervisor deferral helper test pass.
2026-05-03T01:16:42.074632178+00:00 Full cargo test attempted; unrelated integration_chat failures in wg init with no model/route config. Investigating whether env/config can be corrected for validation.
2026-05-03T01:21:17.870198822+00:00 Validation: cargo build passed; cargo install --path . passed. Full cargo test attempted but stopped at pre-existing integration_chat wg init no-model/route failures unrelated to this sentinel change.
2026-05-03T01:21:49.427680280+00:00 Created follow-up fix-update-integration for the unrelated integration_chat wg init no-route failure encountered during full cargo test.
2026-05-03T01:22:14.079179551+00:00 Committed: 60e58824f — pushed to remote
2026-05-03T01:22:27.705377122+00:00 Validated: failing regression tests were added first and failed on missing sentinel API; implementation now makes focused tests pass. Validated: cargo build passes; cargo install --path . passes; git diff --check passes. Full cargo test attempted but blocked by unrelated integration_chat wg init no-route failure; follow-up fix-update-integration created. Live TUI verification not performed in this non-interactive run; ownership boundary is covered by unit tests.
2026-05-03T01:22:36.419756919+00:00 Task pending eval (agent reported done; awaiting `.evaluate-*` to score)
2026-05-03T01:24:59.159443347+00:00 PendingEval → Done (evaluator passed; downstream unblocks)