Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions s08_context_compact/README.ja.md
Original file line number Diff line number Diff line change
Expand Up @@ -85,6 +85,8 @@ for block in ranked:

このステップが対象にするのは、最新のツール結果だけです。完全な出力は保存先から再取得できるため、最初に実行する処理に適しています。

3 万文字は結果ごとの一律の上限ではありません。バッチの合計が 20 万文字以内なら、このステップは 5 万文字の結果も切り詰めません。合計が予算を超えたときだけ、選ばれた大きな結果が次のモデル呼び出し前に保存されます。このステップでは、保存した結果を保存先と先頭 2000 文字に置き換えるため、そのメッセージに末尾の情報は残りません。モデルはツールで保存先を検索するか、必要な短い範囲だけ読み直せます。全文を再取得すると新たな大きな結果が同じ処理を繰り返す可能性があるため、ツールの出力範囲を絞ります。


## ステップ 2:snip_compact

Expand Down
2 changes: 2 additions & 0 deletions s08_context_compact/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -85,6 +85,8 @@ for block in ranked:

This step examines only the latest batch of tool results. The complete output remains available at the saved path, so persistence is the safest operation to run first.

The 30,000-character threshold is not a per-result cap: this step leaves a 50,000-character result intact when its batch stays within 200,000 characters. When the batch exceeds that budget, selected results are saved before the next model call. This step replaces each saved result with its path and first 2,000 characters, so information near the end is absent from that message. The model can search the saved file or request a small excerpt with a tool. Asking for the whole file again may produce another oversized result and repeat the same cycle; request bounded output instead.


## Step 2: snip_compact

Expand Down
2 changes: 2 additions & 0 deletions s08_context_compact/README.zh.md
Original file line number Diff line number Diff line change
Expand Up @@ -85,6 +85,8 @@ for block in ranked:

这一步只处理最新一批工具结果。完整内容仍然可以从路径中取回,因此适合最先执行。

3 万字符不是单条结果的硬上限:如果这一批总量不超过 20 万字符,这一步不会截短一条 5 万字符的结果。只有批次超出预算时,选中的大结果才会在下一次模型调用前转存。这一步用保存路径和前 2000 个字符替换被转存的结果,因此该消息里不再有末尾内容。模型可以用工具搜索保存的文件,或只读取所需的一小段。再次请求整个文件可能产生新的超大结果,反复触发同一处理;应限制工具输出的范围。


## 第二步:snip_compact

Expand Down