[JavaZeroo]

内容

新增 opt-in 的输出消费轨迹 TERMFORGE_OUTPUT_TRACE=<path>：每个被消费的输出块写一行

{"schema":"termforge-output-trace-v1","terminal":1,"t_ms":...,"offset":0,"bytes":347,"columns":111,"rows":7}

• terminal 区分同一场景里的多个窗口；
• offset/bytes 可和 output.bin 对齐到具体字节；
• columns/rows 是消费该块时的网格尺寸。

runner 自动指向场景产物目录的 output-trace.jsonl。

动机（#194）

#194 已确认：resize 期间 ConPTY 的尾部重绘是按旧几何生成、被按新几何消费，错位量 = 高度差（本轮 10−7=3），导致行内容等长互换、选区失效。
要证实"新旧尺寸的先后"，就需要"消费时刻的网格尺寸"这一证据——本 PR 提供的正是它。

本地验证

{"terminal":1,"t_ms":193934578,"offset":0,"bytes":347,"columns":111,"rows":7}
{"terminal":2,"t_ms":193934640,"offset":0,"bytes":105,"columns":74,"rows":17}

边界

• 只增加 opt-in 诊断，不写文件时行为不变；
• 不改变任何 resize/渲染逻辑，不关闭 [R2][调查] 核对 resize 期间 ConPTY 重绘与终端视口/缓冲的对齐 #194。

Refs #194 #156

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-29T17:11:04.200209Z	5dd3a9d	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5dd3a9d5cb

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Record offsets into the shared capture file

In multi-session scenarios, every ShellWindow appends to the same TERMFORGE_CAPTURE_OUTPUT file configured by execute(), but trace_offset_ starts at zero independently for each window. Consequently, the second terminal's first trace entry reports offset 0 even though its bytes follow another terminal's output, and subsequent offsets remain incorrect whenever sessions interleave. This makes the new offset/geometry evidence point at unrelated bytes in output.bin, undermining resize diagnosis. Record the actual shared capture position for each chunk, or give each terminal a separate capture file whose offsets match this counter.

Useful? React with 👍 / 👎.