review-fix-nex-cursor-corruption — Workgraph live mirror

Metadata

Status	abandoned
Agent identity	`3184716484e6f0ea08bb13539daf07686ee79d440505f1fdf2de0357707034c3`
Model	`claude:opus`
Created	2026-05-02T23:57:20.497258432+00:00
Tags	`review,peer-review,nex,chat,quality`, `eval-scheduled`
Failure reason	Consolidating into single review-all-impls task — hit per-agent 10/10 subtask cap; combined review reads all four impl diffs in one session

Description

Cross-model peer review of impl task fix-nex-cursor-corruption (codex:gpt-5.5 worker). Per the user's modulation 2026-05-02 (pattern C): claude:opus reads the impl agent's diff, smoke results, and eval/FLIP scores, then emits an independent verdict.

Impl agent runs on codex:gpt-5.5; this reviewer runs on claude:opus to give cross-model signal.

What to read

The impl's commit(s) on its worktree branch: git log --oneline main..wg/<agent-id>/fix-nex-cursor-corruption then git diff main..wg/<agent-id>/fix-nex-cursor-corruption
The impl task's smoke gate result: wg show fix-nex-cursor-corruption (Validation checklist + log entries)
The impl task's eval/FLIP scores: wg show fix-nex-cursor-corruption (Evaluations section)
The impl task's test output if logged: wg log fix-nex-cursor-corruption

What to produce (via wg log)

A verdict in ONE of two forms:

Form A — concur

VERDICT: concur
Rationale: <2-4 sentences on why the diff + tests + scores look correct as a unit>

Form B — concerns

VERDICT: concerns
Items:
  - <file:line> — <specific issue>
  - <file:line> — <specific issue>
Rationale: <why these matter; whether they block integration or are follow-ups>

Operating constraints

READ ONLY — do not modify source. Your output is a verdict log entry.
Independence — do NOT read the impl agent's own self-assessment before forming your verdict (read it AFTER, only to flag a calibration delta if any)
Calibrated — if the eval/FLIP gave a high score and you concur, say so; if you disagree with the eval verdict itself (not just the impl), flag that as a separate concern

Validation

Verdict logged in correct form
Diff actually read (cite at least 2 file:line locations)
Smoke results actually checked (cite the scenario name + pass/fail)
Eval/FLIP scores actually checked
No source modifications

## Description
Cross-model peer review of impl task `fix-nex-cursor-corruption` (codex:gpt-5.5 worker). Per the user's modulation 2026-05-02 (pattern C): claude:opus reads the impl agent's diff, smoke results, and eval/FLIP scores, then emits an independent verdict.

Impl agent runs on codex:gpt-5.5; this reviewer runs on claude:opus to give cross-model signal.

## What to read
- The impl's commit(s) on its worktree branch: `git log --oneline main..wg/<agent-id>/fix-nex-cursor-corruption` then `git diff main..wg/<agent-id>/fix-nex-cursor-corruption`
- The impl task's smoke gate result: `wg show fix-nex-cursor-corruption` (Validation checklist + log entries)
- The impl task's eval/FLIP scores: `wg show fix-nex-cursor-corruption` (Evaluations section)
- The impl task's test output if logged: `wg log fix-nex-cursor-corruption`

## What to produce (via wg log)
A verdict in ONE of two forms:

### Form A — concur
```
VERDICT: concur
Rationale: <2-4 sentences on why the diff + tests + scores look correct as a unit>
```

### Form B — concerns
```
VERDICT: concerns
Items:
  - <file:line> — <specific issue>
  - <file:line> — <specific issue>
Rationale: <why these matter; whether they block integration or are follow-ups>
```

## Operating constraints
- READ ONLY — do not modify source. Your output is a verdict log entry.
- Independence — do NOT read the impl agent's own self-assessment before forming your verdict (read it AFTER, only to flag a calibration delta if any)
- Calibrated — if the eval/FLIP gave a high score and you concur, say so; if you disagree with the eval verdict itself (not just the impl), flag that as a separate concern

## Validation
- [ ] Verdict logged in correct form
- [ ] Diff actually read (cite at least 2 file:line locations)
- [ ] Smoke results actually checked (cite the scenario name + pass/fail)
- [ ] Eval/FLIP scores actually checked
- [ ] No source modifications

Depends on

Required by

(none)

Log

2026-05-02T23:57:20.479974158+00:00 Task paused
2026-05-02T23:57:41.993261187+00:00 Task abandoned: Consolidating into single review-all-impls task — hit per-agent 10/10 subtask cap; combined review reads all four impl diffs in one session
2026-05-03T00:51:27.907169841+00:00 Task published